ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-04 21:51:28 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK Tile] Int8 Support on CK Tile GEMM (#2267)

Browse Source 
* updates to support int8 in 03_gemm example

* added comments, using aliases, helper functions

* test(gemm_universal): add test cases for int8 gemm pipeline

* fix(test_gemm): fix for failing test unit test for int8

* test(ck_tile): add int8 unit test for gemm universal

* refactor(gemm_universal): GPU reference verification for GEMM code improved

* style(gemm_universal): removed extra comments and did clang format

* merging recent changes to universal gemm to tile_engine

* ck tile engine integration work

* feat(tile_engine): add int8 support to tile engine ops/gemm

* feat(tile_engine): added 32 32 16 mfma instances to tile engine for int8

* style: Format code with clang-format-12

* refactor(tile_engine): address review comments

* style: removed unhelpful comments & unused variables.

* build: tile engine uses default config

* feat: add int8 support for CK_TILE GEMM

* style: added trailing commas to codegen_utils.py

* refactor: tile engine

* refactor: formatting and code review

* refactor: code formatting for python files

* fix: suppress build warning

* add support for gfx950

* refactor:KWarpTile size in gemms util

* Fix the branch and wrap up the k warp tile

* Add bf8 integration

* refactor: clang format and rebase

---------

Co-authored-by: zjli2013 <leezhengjiang@gmail.com>
Co-authored-by: AviralGoelAMD <aviral.goel@amd.com>
Co-authored-by: Khushbu Agarwal <khuagarw@amd.com>
This commit is contained in:
Thomas Ning
2025-06-25 08:20:35 -07:00
committed by GitHub
parent 6d6f4c76c1
commit e03293ebce
24 changed files with 815 additions and 301 deletions
Download Patch File Download Diff File Expand all files Collapse all files

53
example/ck_tile/03_gemm/gemm_utils.hpp
View File

@@ -1,4 +1,3 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
@@ -16,6 +15,25 @@
#define CK_TILE_PIPELINE_COMPUTE_V4 3
#define CK_TILE_PIPELINE_COMPUTE_V5 4
// temporary workaround to get k_warp_tile based on PrecType and gfx950 or not
template <typename PrecType, ck_tile::index_t M_Warp_Tile>
constexpr ck_tile::index_t get_k_warp_tile()
{
#if defined(__gfx950__)
constexpr bool is_8bit_float =
std::is_same_v<PrecType, ck_tile::fp8_t> || std::is_same_v<PrecType, ck_tile::bf8_t>;
if constexpr(M_Warp_Tile == 32)
return is_8bit_float ? 64 : 16;
else
return is_8bit_float ? 128 : 32;
#else
if constexpr(M_Warp_Tile == 32)
return 16;
else
return 32;
#endif
}
struct GemmConfigBase
{
static constexpr bool kPadM = false;
@@ -90,7 +108,7 @@ struct GemmConfigComputeV3 : public GemmConfigBase
static constexpr ck_tile::index_t M_Warp_Tile = 32;
static constexpr ck_tile::index_t N_Warp_Tile = 32;
static constexpr ck_tile::index_t K_Warp_Tile = sizeof(PrecType) == 2 ? 16 : 64;
static constexpr ck_tile::index_t K_Warp_Tile = get_k_warp_tile<PrecType, M_Warp_Tile>();
static constexpr bool DoubleSmemBuffer = false;
static constexpr ck_tile::index_t Pipeline = CK_TILE_PIPELINE_COMPUTE_V3;
@@ -109,7 +127,7 @@ struct GemmConfigComputeV3_1 : public GemmConfigBase
static constexpr ck_tile::index_t M_Warp_Tile = 32;
static constexpr ck_tile::index_t N_Warp_Tile = 32;
static constexpr ck_tile::index_t K_Warp_Tile = sizeof(PrecType) == 2 ? 16 : 64;
static constexpr ck_tile::index_t K_Warp_Tile = get_k_warp_tile<PrecType, M_Warp_Tile>();
static constexpr bool DoubleSmemBuffer = false;
static constexpr ck_tile::index_t Pipeline = CK_TILE_PIPELINE_COMPUTE_V3;
@@ -128,7 +146,7 @@ struct GemmConfigComputeV3_2 : public GemmConfigBase
static constexpr ck_tile::index_t M_Warp_Tile = 16;
static constexpr ck_tile::index_t N_Warp_Tile = 16;
static constexpr ck_tile::index_t K_Warp_Tile = sizeof(PrecType) == 2 ? 32 : 128;
static constexpr ck_tile::index_t K_Warp_Tile = get_k_warp_tile<PrecType, M_Warp_Tile>();
static constexpr bool DoubleSmemBuffer = false;
static constexpr ck_tile::index_t Pipeline = CK_TILE_PIPELINE_COMPUTE_V3;
@@ -151,7 +169,7 @@ struct GemmConfigComputeV4 : public GemmConfigBase
static constexpr ck_tile::index_t M_Warp_Tile = 32;
static constexpr ck_tile::index_t N_Warp_Tile = 32;
static constexpr ck_tile::index_t K_Warp_Tile = sizeof(PrecType) == 2 ? 16 : 64;
static constexpr ck_tile::index_t K_Warp_Tile = get_k_warp_tile<PrecType, M_Warp_Tile>();
static constexpr bool DoubleSmemBuffer = true;
static constexpr ck_tile::index_t Pipeline = CK_TILE_PIPELINE_COMPUTE_V4;
@@ -170,7 +188,7 @@ struct GemmConfigComputeV4_1 : public GemmConfigBase
static constexpr ck_tile::index_t M_Warp_Tile = 32;
static constexpr ck_tile::index_t N_Warp_Tile = 32;
static constexpr ck_tile::index_t K_Warp_Tile = sizeof(PrecType) == 2 ? 16 : 64;
static constexpr ck_tile::index_t K_Warp_Tile = get_k_warp_tile<PrecType, M_Warp_Tile>();
static constexpr bool DoubleSmemBuffer = true;
static constexpr ck_tile::index_t Pipeline = CK_TILE_PIPELINE_COMPUTE_V4;
@@ -189,7 +207,7 @@ struct GemmConfigComputeV5 : public GemmConfigBase
static constexpr ck_tile::index_t M_Warp_Tile = 32;
static constexpr ck_tile::index_t N_Warp_Tile = 32;
static constexpr ck_tile::index_t K_Warp_Tile = sizeof(PrecType) == 2 ? 16 : 64;
static constexpr ck_tile::index_t K_Warp_Tile = get_k_warp_tile<PrecType, M_Warp_Tile>();
static constexpr bool DoubleSmemBuffer = false;
static constexpr ck_tile::index_t Pipeline = CK_TILE_PIPELINE_COMPUTE_V5;
@@ -245,6 +263,15 @@ struct GemmTypeConfig<ck_tile::half_t, ck_tile::pk_int4_t, ck_tile::half_t>
using CDataType = ck_tile::half_t;
};
template <>
struct GemmTypeConfig<ck_tile::int8_t, ck_tile::int8_t, int32_t>
{
using ADataType = ck_tile::int8_t;
using BDataType = ck_tile::int8_t;
using AccDataType = int32_t;
using CDataType = int32_t;
};
template <typename T>
struct DataTypeTraits;
@@ -260,6 +287,12 @@ struct DataTypeTraits<double>
static constexpr const char* name = "fp64";
};
template <>
struct DataTypeTraits<int32_t>
{
static constexpr const char* name = "int32";
};
template <>
struct DataTypeTraits<ck_tile::half_t>
{
@@ -290,6 +323,12 @@ struct DataTypeTraits<ck_tile::pk_int4_t>
static constexpr const char* name = "pk_int4_t";
};
template <>
struct DataTypeTraits<ck_tile::int8_t>
{
static constexpr const char* name = "int8";
};
template <ck_tile::index_t PipelineId>
struct PipelineTypeTraits;