Skip to content

GitLab

Commit eb06c923 authored by Andriy Roshchenko's avatar Andriy Roshchenko
Browse files

Add tests for MFMA_F8F6F4::F32_16x16x128 and MFMA_F8F6F4::F32_32x32x64 instructions

parent a619e3f5
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 563 additions and 154 deletions
CMakeLists.txt
View file @ eb06c923
...@@ -530,7 +530,7 @@ endif() ...@@ -530,7 +530,7 @@ endif()
message("CMAKE_CXX_FLAGS: ${CMAKE_CXX_FLAGS}") message("CMAKE_CXX_FLAGS: ${CMAKE_CXX_FLAGS}")
if("${CMAKE_CXX_COMPILER_ID}" MATCHES "Clang") if("${CMAKE_CXX_COMPILER_ID}" MATCHES "Clang")
add_compile_options(-fcolor-diagnostics) # add_compile_options(-fcolor-diagnostics)
endif() endif()
if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU" AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 4.9) if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU" AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 4.9)
add_compile_options(-fdiagnostics-color=always) add_compile_options(-fdiagnostics-color=always)
......
CMakePresets.json 0 → 100644
View file @ eb06c923
{
"version": 3,
"configurePresets": [
{
"name": "linux-debug",
"displayName": "Linux Debug",
"hidden": true,
"generator": "Unix Makefiles",
"binaryDir": "${sourceDir}/build/${presetName}",
"installDir": "${sourceDir}/build/install/${presetName}",
"environment": {
"MY_ENVIRONMENT_VARIABLE": "NONE",
"PATH": "/usr/local/.cargo/bin:$penv{PATH}",
"SCCACHE_IDLE_TIMEOUT": "11000"
},
"cacheVariables": {
"CMAKE_BUILD_TYPE": "Debug",
"CMAKE_EXPORT_COMPILE_COMMANDS": "ON",
"BUILD_DEV": "ON",
"CMAKE_CXX_COMPILER": "/opt/rocm/bin/hipcc",
"CMAKE_PREFIX_PATH": "/opt/rocm",
"CMAKE_CXX_COMPILER_LAUNCHER": "sccache",
"CMAKE_C_COMPILER_LAUNCHER": "sccache"
},
"condition": {
"type": "equals",
"lhs": "${hostSystemName}",
"rhs": "Linux"
}
},
{
"name": "MI355-debug",
"displayName": "MI355 Debug",
"inherits": "linux-debug",
"description": "Development Environment for MI355.",
"cacheVariables": {
"GPU_TARGETS": "gfx950",
"CMAKE_BUILD_TYPE": "Debug",
"CMAKE_CXX_FLAGS": "-O0 -ggdb"
}
},
{
"name": "MI355-release",
"displayName": "MI355 Release",
"inherits": "linux-debug",
"cacheVariables": {
"GPU_TARGETS": "gfx950",
"CMAKE_BUILD_TYPE": "Release",
"CMAKE_CXX_FLAGS": "-O3"
}
},
{
"name": "MI300X-release",
"displayName": "MI300X Release",
"inherits": "linux-debug",
"cacheVariables": {
"GPU_TARGETS": "gfx942",
"CMAKE_BUILD_TYPE": "Release",
"CMAKE_CXX_FLAGS": "-O3"
}
},
{
"name": "MI250-release",
"displayName": "MI250 Release",
"inherits": "linux-debug",
"cacheVariables": {
"GPU_TARGETS": "gfx90a",
"CMAKE_BUILD_TYPE": "Release",
"CMAKE_CXX_FLAGS": "-O3",
"CK_USE_FP8_ON_UNSUPPORTED_ARCH":"ON"
}
},
{
"name": "MI250-debug",
"displayName": "MI250 Debug",
"inherits": "linux-debug",
"cacheVariables": {
"GPU_TARGETS": "gfx90a",
"CMAKE_BUILD_TYPE": "Debug",
"CMAKE_CXX_FLAGS": "-O0 -ggdb",
"CK_USE_FP8_ON_UNSUPPORTED_ARCH":"ON"
}
},
{
"name": "RX7800-release",
"displayName": "RX7800 Release",
"inherits": "linux-debug",
"cacheVariables": {
"GPU_TARGETS": "gfx1101",
"DL_KERNELS": "ON",
"CMAKE_BUILD_TYPE": "Release",
"CMAKE_CXX_FLAGS": "-O3"
}
},
{
"name": "RX7800-debug",
"displayName": "RX7800 Debug",
"inherits": "linux-debug",
"cacheVariables": {
"GPU_TARGETS": "gfx1101",
"DL_KERNELS": "ON",
"CMAKE_BUILD_TYPE": "Debug",
"CMAKE_CXX_FLAGS": "-O0 -ggdb"
}
}
],
"buildPresets": [
{
"name": "Debug",
"hidden": true,
"configuration": "Debug"
},
{
"name": "Release",
"hidden": true,
"configuration": "Release"
},
{
"name": "MI355-debug",
"displayName": "MI355",
"configurePreset": "MI355-debug",
"description": "Build Environment for MI355 Debug.",
"inherits": [
"Debug"
],
"jobs": 128
},
{
"name": "MI355-release",
"displayName": "MI355",
"configurePreset": "MI355-release",
"description": "Build Environment for MI355 Release.",
"inherits": [
"Release"
],
"jobs": 128
},
{
"name": "MI300X-release",
"displayName": "MI300X",
"configurePreset": "MI300X-release",
"description": "Build Environment for MI300X Release.",
"inherits": [
"Release"
],
"jobs": 128
},
{
"name": "MI250-release",
"displayName": "MI250",
"configurePreset": "MI250-release",
"description": "Build Environment for MI250 Release.",
"inherits": [
"Release"
],
"jobs": 128
},
{
"name": "MI250-debug",
"displayName": "MI250",
"configurePreset": "MI250-debug",
"description": "Build Environment for MI250 Debug.",
"inherits": [
"Debug"
],
"jobs": 128
},
{
"name": "RX7800-release",
"displayName": "RX7800",
"configurePreset": "RX7800-release",
"description": "Build Environment for RX7800 Release.",
"inherits": [
"Release"
],
"jobs": 128
},
{
"name": "RX7800-debug",
"displayName": "RX7800",
"configurePreset": "RX7800-debug",
"description": "Build Environment for RX7800 Debug.",
"inherits": [
"Debug"
],
"jobs": 128
}
]
}
test/mx_mfma_op/mx_mfma_op.cpp
View file @ eb06c923
...@@ -6,52 +6,55 @@ ...@@ -6,52 +6,55 @@
#include "mx_mfma_op.hpp" #include "mx_mfma_op.hpp"
using ck::e8m0_bexp_t; using ck::e8m0_bexp_t;
using ck::f8_ocp_t; using ck::f8_t;
using ck::half_t;
using ck::type_convert; using ck::type_convert;
template <typename Src1Type, template <typename AType, typename BType, typename CType, ck::mx_mfma_test::MFMA_F8F6F4 mfma>
ck::index_t Src1VecSize,
typename Src2Type,
ck::index_t Src2VecSize,
typename DstType,
ck::index_t AccVecSize,
typename AccType,
typename CPUAccType,
ck::index_t M,
ck::index_t N,
ck::index_t K>
bool run_test() bool run_test()
{ {
using Row = ck::tensor_layout::gemm::RowMajor; using ALayout = ck::tensor_layout::gemm::ColumnMajor;
using PassThrough = ck::tensor_operation::element_wise::PassThrough; using BLayout = ck::tensor_layout::gemm::ColumnMajor;
bool pass = true; using CLayout = ck::tensor_layout::gemm::ColumnMajor;
const auto mx_mfma_kernel = ck::mx_mfma_test:: using AccType = float; // only MFMA_F32 instructions supported
matmul<Src1Type, Src1VecSize, Src2Type, Src2VecSize, AccType, AccVecSize, DstType, M, N, K>; using CPUAccType = AccType;
pass = ck::mx_mfma_test::TestMXMFMA<decltype(mx_mfma_kernel), ck::mfma_type<static_cast<ck::MfmaInstr>(mfma)> mfma_instr;
Src1Type, constexpr auto BLOCK_M = mfma_instr.m_per_blk;
Src2Type, constexpr auto BLOCK_N = mfma_instr.n_per_blk;
DstType, constexpr auto BLOCK_K = mfma_instr.num_input_blks * mfma_instr.k_per_blk;
AccType,
CPUAccType, const auto mx_mfma_kernel =
decltype(Row{}), ck::mx_mfma_test::matmul<AType, BType, CType, AccType, BLOCK_M, BLOCK_N, BLOCK_K>;
decltype(Row{}),
decltype(Row{}), bool pass = true;
PassThrough,
PassThrough, pass = ck::mx_mfma_test::TestMFMA<decltype(mx_mfma_kernel),
PassThrough, AType,
AccVecSize, BType,
M, CType,
N, AccType,
K>{}(mx_mfma_kernel); CPUAccType,
ALayout,
BLayout,
CLayout,
BLOCK_M,
BLOCK_N,
BLOCK_K>{}(mx_mfma_kernel);
return pass; return pass;
} }
TEST(MXMFMA, FP8MFMA16x16x128) TEST(MFMA, FP8MFMA16x16x128)
{
auto pass = run_test<f8_t, f8_t, half_t, ck::mx_mfma_test::MFMA_F8F6F4::F32_16x16x128>();
EXPECT_TRUE(pass);
}
TEST(MFMA, FP8MFMA32x32x64)
{ {
auto pass = run_test<float, 1, float, 1, float, 1, float, float, 16, 16, 128>(); auto pass = run_test<f8_t, f8_t, float, ck::mx_mfma_test::MFMA_F8F6F4::F32_32x32x64>();
EXPECT_TRUE(pass); EXPECT_TRUE(pass);
} }
...@@ -70,5 +73,5 @@ TEST(MXMFMA, FP8MFMA16x16x128) ...@@ -70,5 +73,5 @@ TEST(MXMFMA, FP8MFMA16x16x128)
// EXPECT_TRUE(run_test<bf8, 1, bf8, 1, float, 1, float, float, 32, 32, 64>()); // EXPECT_TRUE(run_test<bf8, 1, bf8, 1, float, 1, float, float, 32, 32, 64>());
// } // }
TEST(MXMFMA, MXFP8xMXFP8) { EXPECT_TRUE(false) << "Not Implemented\n"; } // TEST(MXMFMA, MXFP8xMXFP8) { EXPECT_TRUE(false) << "Not Implemented\n"; }
TEST(MXMFMA, MXBF8xMXBF8) { EXPECT_TRUE(false) << "Not Implemented\n"; } // TEST(MXMFMA, MXBF8xMXBF8) { EXPECT_TRUE(false) << "Not Implemented\n"; }
test/mx_mfma_op/mx_mfma_op.hpp
View file @ eb06c923
This diff is collapsed.
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