mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-11 17:00:18 +00:00
[Block Scale GEMM] Optimized block scale gemm (#1950)
* Added two kernel for M=32 problem * Comment the first one * Enable multiply_multiply for Scale_Block_M = 1 for deepseek * Modify the a_thread offset since the A data load is different from B. * edit fp8 ab scale for Scale_Block_M=1 * edit GemmSpec to MNKPadding * enable blockwise pipelie v1 and v2. v1 is work for small K. * add instance for gemm_ab_scale * fix cmakelist of ckProfiler * optimize blockscale gemm. todo: reduce vgpr usage * fix a correctness bug * sanity checked * revert ckprofiler cmake changes * clang format * revert unnecessary changes. * remove commented codes. * split weight preshuffle library targets * bring back enable-post-misched=0 * fix build issues for gemm_multiply_multiply_fp8 instances * fix clang format * add verbose build flag when building for all targets * reduce path names for new instances * fix paths in cmake * refactor gemm_multiply_multiply library target * fix a bug in example * fix example 65 cmake * reduce the number of threads when building libs for all targets to 50 * use ninja to build for all targets * reduce teh number of threads when building for all targets * reduce the number of threads to 32 when building libs for all targets to 50 --------- Co-authored-by: mtgu0705 <mtgu@amd.com> Co-authored-by: chenjun <junchen2@amd.com> Co-authored-by: illsilin <Illia.Silin@amd.com> Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
This commit is contained in:
Haocong WANG
@@ -17,7 +17,7 @@ namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace instance {
|
||||
#if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_FP8))
|
||||
void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_default_instances(
|
||||
void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_1_128_128_comp_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScale<Row,
|
||||
Col,
|
||||
Tuple<>,
|
||||
@@ -28,14 +28,14 @@ void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_default_i
|
||||
F32,
|
||||
Tuple<>,
|
||||
BF16,
|
||||
128,
|
||||
1,
|
||||
128,
|
||||
128,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
|
||||
void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_kpadding_instances(
|
||||
void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_1_128_128_comp_kpadding_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScale<Row,
|
||||
Col,
|
||||
Tuple<>,
|
||||
@@ -46,14 +46,14 @@ void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_kpadding_
|
||||
F32,
|
||||
Tuple<>,
|
||||
BF16,
|
||||
128,
|
||||
1,
|
||||
128,
|
||||
128,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
|
||||
void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_mnpadding_instances(
|
||||
void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_1_128_128_mem_v1_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScale<Row,
|
||||
Col,
|
||||
Tuple<>,
|
||||
@@ -64,14 +64,14 @@ void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_mnpadding
|
||||
F32,
|
||||
Tuple<>,
|
||||
BF16,
|
||||
128,
|
||||
1,
|
||||
128,
|
||||
128,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
|
||||
void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_mnkpadding_instances(
|
||||
void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_1_128_128_mem_v1_kpadding_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScale<Row,
|
||||
Col,
|
||||
Tuple<>,
|
||||
@@ -82,61 +82,7 @@ void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_mnkpaddin
|
||||
F32,
|
||||
Tuple<>,
|
||||
BF16,
|
||||
128,
|
||||
128,
|
||||
128,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
|
||||
void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScale<Row,
|
||||
Col,
|
||||
Tuple<>,
|
||||
Row,
|
||||
F8,
|
||||
F32,
|
||||
F8,
|
||||
F32,
|
||||
Tuple<>,
|
||||
BF16,
|
||||
128,
|
||||
128,
|
||||
128,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
|
||||
void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_kpadding_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScale<Row,
|
||||
Col,
|
||||
Tuple<>,
|
||||
Row,
|
||||
F8,
|
||||
F32,
|
||||
F8,
|
||||
F32,
|
||||
Tuple<>,
|
||||
BF16,
|
||||
128,
|
||||
128,
|
||||
128,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
|
||||
void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_mnkpadding_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScale<Row,
|
||||
Col,
|
||||
Tuple<>,
|
||||
Row,
|
||||
F8,
|
||||
F32,
|
||||
F8,
|
||||
F32,
|
||||
Tuple<>,
|
||||
BF16,
|
||||
128,
|
||||
1,
|
||||
128,
|
||||
128,
|
||||
PassThrough,
|
||||
@@ -163,7 +109,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
|
||||
B1DataType,
|
||||
Tuple<>,
|
||||
CDataType,
|
||||
128,
|
||||
1,
|
||||
128,
|
||||
128,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
@@ -180,7 +126,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
|
||||
B1DataType,
|
||||
Tuple<>,
|
||||
CDataType,
|
||||
128,
|
||||
1,
|
||||
128,
|
||||
128,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
@@ -198,20 +144,14 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
|
||||
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
|
||||
is_same_v<CLayout, Row>)
|
||||
{
|
||||
add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_default_instances(
|
||||
add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_1_128_128_comp_default_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_kpadding_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_mnpadding_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_mnkpadding_instances(
|
||||
add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_1_128_128_comp_kpadding_instances(
|
||||
op_ptrs);
|
||||
|
||||
add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_default_instances(
|
||||
add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_1_128_128_mem_v1_default_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_kpadding_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_mnkpadding_instances(
|
||||
add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_1_128_128_mem_v1_kpadding_instances(
|
||||
op_ptrs);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -122,6 +122,32 @@ void add_device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_comp_mfma16x16_kp
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>& instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_comp_mfma16x16_default_instances_part3(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>& instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_comp_mfma16x16_kpadding_instances_part3(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>& instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_mem_v1_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
|
||||
Col,
|
||||
@@ -360,6 +386,32 @@ void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_mfma16x16_kpa
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>& instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_mfma16x16_default_instances_part3(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>& instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_mfma16x16_kpadding_instances_part3(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>& instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v1_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
|
||||
Col,
|
||||
@@ -556,6 +608,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_comp_mfma16x16_kpadding_instances_part2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_comp_mfma16x16_default_instances_part3(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_comp_mfma16x16_kpadding_instances_part3(
|
||||
op_ptrs);
|
||||
|
||||
add_device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_mem_v1_default_instances(
|
||||
op_ptrs);
|
||||
@@ -593,6 +649,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_mfma16x16_kpadding_instances_part2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_mfma16x16_default_instances_part3(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_mfma16x16_kpadding_instances_part3(
|
||||
op_ptrs);
|
||||
|
||||
add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v1_default_instances(
|
||||
op_ptrs);
|
||||
|
||||
@@ -1,317 +0,0 @@
|
||||
// SPDX-License-Identifier: MIT
|
||||
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <vector>
|
||||
#include <memory>
|
||||
#include "ck/ck.hpp"
|
||||
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
|
||||
#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle_v3_b_preshuffle.hpp"
|
||||
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
|
||||
|
||||
#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace instance {
|
||||
|
||||
#if(defined(CK_ENABLE_F16) || defined(CK_ENABLE_FP8))
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p1_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p2_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p3_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p1_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p2_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p3_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_compute_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
#endif
|
||||
|
||||
#if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_FP8))
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p1_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p2_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p3_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p1_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p2_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p3_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_compute_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
#endif
|
||||
|
||||
template <typename ADataType,
|
||||
typename BDataType,
|
||||
typename CDataType,
|
||||
typename ALayout,
|
||||
typename BLayout,
|
||||
typename CLayout>
|
||||
struct DeviceOperationInstanceFactory<
|
||||
ck::tensor_operation::device::DeviceGemmMultipleDSplitKBPreShuffle<
|
||||
ALayout,
|
||||
BLayout,
|
||||
Tuple<Row, Col>,
|
||||
CLayout,
|
||||
ADataType,
|
||||
BDataType,
|
||||
Tuple<F32, F32>,
|
||||
CDataType,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::MultiplyMultiply>>
|
||||
{
|
||||
using DeviceOp =
|
||||
DeviceGemmMultipleDSplitKBPreShuffle<ALayout,
|
||||
BLayout,
|
||||
Tuple<Row, Col>,
|
||||
CLayout,
|
||||
ADataType,
|
||||
BDataType,
|
||||
Tuple<F32, F32>,
|
||||
CDataType,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::MultiplyMultiply>;
|
||||
|
||||
static auto GetInstances()
|
||||
{
|
||||
std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
|
||||
// TODO: Add MFMA layout into tensor layout
|
||||
#if(defined(CK_ENABLE_F16) || defined(CK_ENABLE_FP8))
|
||||
if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, f8_t> &&
|
||||
is_same_v<CDataType, half_t>)
|
||||
{
|
||||
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
|
||||
is_same_v<CLayout, Row>)
|
||||
{
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p1_default_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p2_default_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p3_default_instances(
|
||||
op_ptrs);
|
||||
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p1_default_instances_v2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p2_default_instances_v2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p3_default_instances_v2(
|
||||
op_ptrs);
|
||||
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_compute_default_instances(
|
||||
op_ptrs);
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
#if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_FP8))
|
||||
if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, f8_t> &&
|
||||
is_same_v<CDataType, bhalf_t>)
|
||||
{
|
||||
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
|
||||
is_same_v<CLayout, Row>)
|
||||
{
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p1_default_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p2_default_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p3_default_instances(
|
||||
op_ptrs);
|
||||
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p1_default_instances_v2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p2_default_instances_v2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p3_default_instances_v2(
|
||||
op_ptrs);
|
||||
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_compute_default_instances(
|
||||
op_ptrs);
|
||||
}
|
||||
}
|
||||
#endif
|
||||
return op_ptrs;
|
||||
}
|
||||
};
|
||||
|
||||
} // namespace instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
@@ -0,0 +1,664 @@
|
||||
// SPDX-License-Identifier: MIT
|
||||
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <vector>
|
||||
#include <memory>
|
||||
#include "ck/ck.hpp"
|
||||
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
|
||||
#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle_v3_b_preshuffle.hpp"
|
||||
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
|
||||
|
||||
#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace instance {
|
||||
|
||||
#if(defined(CK_ENABLE_F16) || defined(CK_ENABLE_FP8))
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p1_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p2_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p3_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p4_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p5_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p1_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p2_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p3_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p4_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p5_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_compute_default_instances_p1(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_compute_default_instances_p2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma16x16_mn_compute_default_instances_p1(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma16x16_mn_compute_default_instances_p2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma16x16_mn_compute_default_instances_p3(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma16x16_mn_compute_default_instances_p4(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma16x16_mn_compute_default_instances_p5(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma16x16_mn_compute_default_instances_p6(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
#endif
|
||||
|
||||
#if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_FP8))
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p1_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p2_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p3_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p4_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p5_default_instances(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p1_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p2_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p3_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p4_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p5_default_instances_v2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_compute_default_instances_p1(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_compute_default_instances_p2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma16x16_mn_compute_default_instances_p1(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma16x16_mn_compute_default_instances_p2(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma16x16_mn_compute_default_instances_p3(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma16x16_mn_compute_default_instances_p4(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma16x16_mn_compute_default_instances_p5(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
void add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma16x16_mn_compute_default_instances_p6(
|
||||
std::vector<std::unique_ptr<DeviceGemmMultipleDSplitKBPreShuffle<Row,
|
||||
Col,
|
||||
Tuple<Row, Col>,
|
||||
Row,
|
||||
F8,
|
||||
F8,
|
||||
Tuple<F32, F32>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
MultiplyMultiply>>>&
|
||||
instances);
|
||||
|
||||
#endif
|
||||
|
||||
template <typename ADataType,
|
||||
typename BDataType,
|
||||
typename CDataType,
|
||||
typename ALayout,
|
||||
typename BLayout,
|
||||
typename CLayout>
|
||||
struct DeviceOperationInstanceFactory<
|
||||
ck::tensor_operation::device::DeviceGemmMultipleDSplitKBPreShuffle<
|
||||
ALayout,
|
||||
BLayout,
|
||||
Tuple<Row, Col>,
|
||||
CLayout,
|
||||
ADataType,
|
||||
BDataType,
|
||||
Tuple<F32, F32>,
|
||||
CDataType,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::MultiplyMultiply>>
|
||||
{
|
||||
using DeviceOp =
|
||||
DeviceGemmMultipleDSplitKBPreShuffle<ALayout,
|
||||
BLayout,
|
||||
Tuple<Row, Col>,
|
||||
CLayout,
|
||||
ADataType,
|
||||
BDataType,
|
||||
Tuple<F32, F32>,
|
||||
CDataType,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::MultiplyMultiply>;
|
||||
|
||||
static auto GetInstances()
|
||||
{
|
||||
std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
|
||||
// TODO: Add MFMA layout into tensor layout
|
||||
#if(defined(CK_ENABLE_F16) || defined(CK_ENABLE_FP8))
|
||||
if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, f8_t> &&
|
||||
is_same_v<CDataType, half_t>)
|
||||
{
|
||||
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
|
||||
is_same_v<CLayout, Row>)
|
||||
{
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p1_default_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p2_default_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p3_default_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p4_default_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p5_default_instances(
|
||||
op_ptrs);
|
||||
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p1_default_instances_v2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p2_default_instances_v2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p3_default_instances_v2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p4_default_instances_v2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p5_default_instances_v2(
|
||||
op_ptrs);
|
||||
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_compute_default_instances_p1(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_compute_default_instances_p2(
|
||||
op_ptrs);
|
||||
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma16x16_mn_compute_default_instances_p1(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma16x16_mn_compute_default_instances_p2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma16x16_mn_compute_default_instances_p3(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma16x16_mn_compute_default_instances_p4(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma16x16_mn_compute_default_instances_p5(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_f16_mk_mfma16x16_mn_compute_default_instances_p6(
|
||||
op_ptrs);
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
#if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_FP8))
|
||||
if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, f8_t> &&
|
||||
is_same_v<CDataType, bhalf_t>)
|
||||
{
|
||||
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
|
||||
is_same_v<CLayout, Row>)
|
||||
{
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p1_default_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p2_default_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p3_default_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p4_default_instances(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p5_default_instances(
|
||||
op_ptrs);
|
||||
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p1_default_instances_v2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p2_default_instances_v2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p3_default_instances_v2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p4_default_instances_v2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_p5_default_instances_v2(
|
||||
op_ptrs);
|
||||
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_compute_default_instances_p1(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma_mn_compute_default_instances_p2(
|
||||
op_ptrs);
|
||||
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma16x16_mn_compute_default_instances_p1(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma16x16_mn_compute_default_instances_p2(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma16x16_mn_compute_default_instances_p3(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma16x16_mn_compute_default_instances_p4(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma16x16_mn_compute_default_instances_p5(
|
||||
op_ptrs);
|
||||
add_device_gemm_multiply_multiply_weight_preshuffle_xdl_f8_f8_bf16_mk_mfma16x16_mn_compute_default_instances_p6(
|
||||
op_ptrs);
|
||||
}
|
||||
}
|
||||
#endif
|
||||
return op_ptrs;
|
||||
}
|
||||
};
|
||||
|
||||
} // namespace instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
Reference in New Issue
Block a user