Skip to content

Commit ab39002

Browse files
author
Yakov Lipkovich
authored
Implement MessagePack backend (ROCm#904)
* Implement MessagePack backend * Install msgpack on Docker by adding Debian repo and installing from there * Revert "Install msgpack on Docker by adding Debian repo and installing from there" This reverts commit 94ae9de.
1 parent a93bc68 commit ab39002

38 files changed

+701
-188
lines changed

Diff for: ‎HostLibraryTests/ContractionLibraryLoading_test.cpp

+3-3
Original file line numberDiff line numberDiff line change
@@ -37,7 +37,7 @@ using namespace Tensile;
3737
TEST(ContractionLibraryLoadingTest, MultipleKernels)
3838
{
3939
auto library = LoadLibraryFile<ContractionProblem>(
40-
TestData::Instance().file("SampleTensileKernels.yaml").native());
40+
TestData::Instance().file("SampleTensileKernels").native());
4141
ASSERT_NE(library, nullptr);
4242

4343
AMDGPU hardware;
@@ -87,7 +87,7 @@ TEST(ContractionLibraryLoadingTest, MultipleKernels)
8787

8888
TEST(ContractionLibraryLoadingTest, SGEMM_Kernels_Lite)
8989
{
90-
auto library = LoadLibraryFile<ContractionProblem>(
91-
TestData::Instance().file("KernelsLite.yaml").native());
90+
auto library
91+
= LoadLibraryFile<ContractionProblem>(TestData::Instance().file("KernelsLite").native());
9292
ASSERT_NE(library, nullptr);
9393
}

Diff for: ‎HostLibraryTests/TestData_test.cpp

+2-2
Original file line numberDiff line numberDiff line change
@@ -37,10 +37,10 @@ TEST(TestData, Simple)
3737
auto is_regular_file
3838
= static_cast<bool (*)(boost::filesystem::path const&)>(boost::filesystem::is_regular_file);
3939

40-
EXPECT_PRED1(is_regular_file, data.file("KernelsLite.yaml"));
40+
EXPECT_PRED1(is_regular_file, data.file("KernelsLite"));
4141
EXPECT_FALSE(is_regular_file(data.file("fjdlksljfjldskj")));
4242

43-
auto files = data.glob("*.yaml");
43+
auto files = data.glob(std::string("*.") + TestData::defaultExtension);
4444
EXPECT_EQ(files.size(), 6);
4545
for(auto file : files)
4646
{

Diff for: ‎HostLibraryTests/configs/CMakeLists.txt

+21-11
Original file line numberDiff line numberDiff line change
@@ -1,11 +1,21 @@
1-
2-
set(SOLUTION_LIBRARY_FILES
3-
${SOLUTION_LIBRARY_FILES}
4-
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/KernelsLite.yaml"
5-
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/KernelsLiteMixed.yaml"
6-
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/KernelsLiteNavi.yaml"
7-
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/KernelsTileLite.yaml"
8-
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/SampleTensileKernels.yaml"
9-
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/rocBLAS_Full.yaml"
10-
PARENT_SCOPE)
11-
1+
if(Tensile_YAML)
2+
set(SOLUTION_LIBRARY_FILES
3+
${SOLUTION_LIBRARY_FILES}
4+
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/KernelsLite.yaml"
5+
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/KernelsLiteMixed.yaml"
6+
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/KernelsLiteNavi.yaml"
7+
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/KernelsTileLite.yaml"
8+
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/SampleTensileKernels.yaml"
9+
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/rocBLAS_Full.yaml"
10+
PARENT_SCOPE)
11+
else()
12+
set(SOLUTION_LIBRARY_FILES
13+
${SOLUTION_LIBRARY_FILES}
14+
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/KernelsLite.dat"
15+
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/KernelsLiteMixed.dat"
16+
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/KernelsLiteNavi.dat"
17+
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/KernelsTileLite.dat"
18+
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/SampleTensileKernels.dat"
19+
"${CMAKE_CURRENT_SOURCE_DIR}/SolutionLibraries/rocBLAS_Full.dat"
20+
PARENT_SCOPE)
21+
endif()

Diff for: ‎HostLibraryTests/configs/SolutionLibraries/KernelsLite.dat

897 KB
Binary file not shown.

Diff for: ‎HostLibraryTests/configs/SolutionLibraries/KernelsLiteMixed.dat

1.42 MB
Binary file not shown.

Diff for: ‎HostLibraryTests/configs/SolutionLibraries/KernelsLiteNavi.dat

881 KB
Binary file not shown.

Diff for: ‎HostLibraryTests/configs/SolutionLibraries/KernelsTileLite.dat

290 KB
Binary file not shown.

Diff for: ‎HostLibraryTests/configs/SolutionLibraries/SampleTensileKernels.dat

2.02 KB
Binary file not shown.

Diff for: ‎HostLibraryTests/configs/SolutionLibraries/rocBLAS_Full.dat

54.3 MB
Binary file not shown.

Diff for: ‎HostLibraryTests/hip/RunGEMMKernelTileSelection_test.cpp

+15-12
Original file line numberDiff line numberDiff line change
@@ -176,17 +176,20 @@ struct RunGEMMKernelSolutionSelectionTest : public ::testing::TestWithParam<Cont
176176
#endif
177177
}
178178

179-
void TearDown() override
180-
{
181-
hipFree(a_d);
182-
hipFree(b_d);
183-
hipFree(c_d);
184-
hipFree(d_d);
185-
hipFree(d_ref_d);
179+
}
186180

187-
hipDeviceReset();
188-
}
189-
};
181+
void TearDown() override
182+
{
183+
hipFree(a_d);
184+
hipFree(b_d);
185+
hipFree(c_d);
186+
hipFree(d_d);
187+
hipFree(d_ref_d);
188+
189+
hipDeviceReset();
190+
}
191+
}
192+
;
190193

191194
TEST_P(RunGEMMKernelSolutionSelectionTest, KernelsTileSelection)
192195
{
@@ -195,11 +198,11 @@ TEST_P(RunGEMMKernelSolutionSelectionTest, KernelsTileSelection)
195198
// std::cout << problem << std::endl;
196199

197200
auto library = LoadLibraryFile<ContractionProblem>(
198-
TestData::File("kernels_tile_selection/TensileLibrary.yaml").native());
201+
TestData::Instance().file("kernels_tile_selection/TensileLibrary").native());
199202

200203
hip::SolutionAdapter adapter(false);
201204
adapter.loadCodeObjectFile(
202-
TestData::File("kernels_tile_selection/TensileLibrary_gfx906.co").native());
205+
TestData::Instance().file("kernels_tile_selection/TensileLibrary_gfx906", "co").native());
203206

204207
ASSERT_NE(library, nullptr);
205208

Diff for: ‎HostLibraryTests/hip/RunGEMMKernel_test.cpp

+4-5
Original file line numberDiff line numberDiff line change
@@ -534,7 +534,7 @@ std::vector<std::tuple<std::shared_ptr<SolutionLibrary<ContractionProblem>>,
534534

535535
{
536536
auto library = LoadLibraryFile<ContractionProblem>(
537-
TestData::Instance().file("kernels_lite/TensileLibrary.yaml").native());
537+
TestData::Instance().file("kernels_lite/TensileLibrary").native());
538538
auto adapter = std::make_shared<hip::SolutionAdapter>(debug, "kernels_lite (file)");
539539
for(auto file : TestData::Instance().glob("kernels_lite/*.*co"))
540540
adapter->loadCodeObjectFile(file.native());
@@ -544,7 +544,7 @@ std::vector<std::tuple<std::shared_ptr<SolutionLibrary<ContractionProblem>>,
544544

545545
{
546546
auto library = LoadLibraryFile<ContractionProblem>(
547-
TestData::Instance().file("kernels_lite_mixed/TensileLibrary.yaml").native());
547+
TestData::Instance().file("kernels_lite_mixed/TensileLibrary").native());
548548
auto adapter = std::make_shared<hip::SolutionAdapter>(debug, "kernels_lite_mixed (file)");
549549
for(auto file : TestData::Instance().glob("kernels_lite_mixed/*.*co"))
550550
adapter->loadCodeObjectFile(file.native());
@@ -554,7 +554,7 @@ std::vector<std::tuple<std::shared_ptr<SolutionLibrary<ContractionProblem>>,
554554

555555
{
556556
auto library = LoadLibraryFile<ContractionProblem>(
557-
TestData::Instance().file("tile_aware_selection/library/TensileLibrary.yaml").native());
557+
TestData::Instance().file("tile_aware_selection/library/TensileLibrary").native());
558558

559559
auto adapter = std::make_shared<hip::SolutionAdapter>(debug, "tile_aware_selection");
560560
for(auto file : TestData::Instance().glob("tile_aware_selection/library/*.*co"))
@@ -569,8 +569,7 @@ std::vector<std::tuple<std::shared_ptr<SolutionLibrary<ContractionProblem>>,
569569
auto envDir = TestData::Env("TENSILE_TEST_LIBRARY");
570570
if(envDir)
571571
{
572-
auto library
573-
= LoadLibraryFile<ContractionProblem>(envDir.file("TensileLibrary.yaml").native());
572+
auto library = LoadLibraryFile<ContractionProblem>(envDir.file("TensileLibrary").native());
574573
auto adapter = std::make_shared<hip::SolutionAdapter>(debug, "TENSILE_TEST_LIBRARY");
575574

576575
for(auto file : envDir.glob("*.co"))

Diff for: ‎HostLibraryTests/llvm/LibraryPerformance_test.cpp

+15-15
Original file line numberDiff line numberDiff line change
@@ -37,16 +37,16 @@ using namespace Tensile;
3737
/**
3838
* LibraryPerformanceTest:
3939
*
40-
* This suite contains micro-benchmarks for pieces of the runtime library. It
41-
* does not exercise any of the Hip-specific code.
40+
* This suite contains micro-benchmarks for pieces of the runtime library. It does not
41+
* exercise any of the Hip-specific code.
4242
*
43-
* There are no performance-based assertions or checks. The timing results are
44-
* provided by googletest.
43+
* There are no performance-based assertions or checks. The timing results are provided by
44+
* googletest.
4545
*
46-
* Most of these tests depend on a library being loaded from a YAML file. The
47-
* library objects are cached so that the deserialization time is not a part of
48-
* the actual test (outside of the LoadLibrary test). PopulateCache is an empty
49-
* test whose purpose is to ensure the cache is populated for the actual tests.
46+
* Most of these tests depend on a library being loaded from a DAT/YAML file. The library objects
47+
* are cached so that the deserialization time is not a part of the actual test (outside of the
48+
* LoadLibrary test). PopulateCache is an empty test whose purpose is to ensure the cache is
49+
* populated for the actual tests.
5050
*/
5151
struct LibraryPerformanceTest
5252
: public ::testing::TestWithParam<std::tuple<AMDGPU, std::string, bool, bool>>
@@ -223,17 +223,17 @@ std::vector<LibraryPerformanceTest::ParamType> GetParams()
223223

224224
for(auto const& gpu : gpus)
225225
{
226-
rv.push_back(std::make_tuple(gpu, "KernelsLite.yaml", false, false));
227-
rv.push_back(std::make_tuple(gpu, "KernelsLiteMixed.yaml", false, true));
228-
rv.push_back(std::make_tuple(gpu, "KernelsLiteNavi.yaml", true, false));
229-
rv.push_back(std::make_tuple(gpu, "KernelsTileLite.yaml", false, false));
230-
rv.push_back(std::make_tuple(gpu, "rocBLAS_Full.yaml", false, true));
226+
rv.push_back(std::make_tuple(gpu, "KernelsLite", false, false));
227+
rv.push_back(std::make_tuple(gpu, "KernelsLiteMixed", false, true));
228+
rv.push_back(std::make_tuple(gpu, "KernelsLiteNavi", true, false));
229+
rv.push_back(std::make_tuple(gpu, "KernelsTileLite", false, false));
230+
rv.push_back(std::make_tuple(gpu, "rocBLAS_Full", false, true));
231231
}
232232

233233
rv.push_back(std::make_tuple(
234-
AMDGPU(AMDGPU::Processor::gfx908, 64, "Arcturus"), "rocBLAS_Full.yaml", false, true));
234+
AMDGPU(AMDGPU::Processor::gfx908, 64, "Arcturus"), "rocBLAS_Full", false, true));
235235
rv.push_back(std::make_tuple(
236-
AMDGPU(AMDGPU::Processor::gfx1010, 40, "Navi"), "KernelsLiteNavi.yaml", true, true));
236+
AMDGPU(AMDGPU::Processor::gfx1010, 40, "Navi"), "KernelsLiteNavi", true, true));
237237

238238
return rv;
239239
}

Diff for: ‎HostLibraryTests/testlib/include/TestData.hpp

+3-1
Original file line numberDiff line numberDiff line change
@@ -41,7 +41,9 @@ struct TestData : public Tensile::LazySingleton<TestData>
4141

4242
boost::filesystem::path dataDir() const;
4343

44-
boost::filesystem::path file(std::string const& filename) const;
44+
static const std::string defaultExtension;
45+
boost::filesystem::path file(std::string const& filename,
46+
std::string const& extension = defaultExtension) const;
4547

4648
std::vector<boost::filesystem::path> glob(std::string const& pattern) const;
4749

Diff for: ‎HostLibraryTests/testlib/source/TestData.cpp

+9-2
Original file line numberDiff line numberDiff line change
@@ -39,6 +39,12 @@
3939

4040
#include <Tensile/Utils.hpp>
4141

42+
#ifdef TENSILE_YAML
43+
const std::string TestData::defaultExtension = "yaml";
44+
#else
45+
const std::string TestData::defaultExtension = "dat";
46+
#endif
47+
4248
TestData::operator bool() const
4349
{
4450
return boost::filesystem::is_directory(dataDir());
@@ -64,9 +70,10 @@ boost::filesystem::path TestData::dataDir() const
6470
return m_dataDir;
6571
}
6672

67-
boost::filesystem::path TestData::file(std::string const& filename) const
73+
boost::filesystem::path TestData::file(std::string const& filename,
74+
std::string const& extension) const
6875
{
69-
return dataDir() / filename;
76+
return dataDir() / (filename + "." + extension);
7077
}
7178

7279
std::vector<boost::filesystem::path> TestData::glob(std::string const& pattern) const

Diff for: ‎Tensile/BenchmarkProblems.py

+5-4
Original file line numberDiff line numberDiff line change
@@ -32,7 +32,7 @@
3232

3333
from . import ClientExecutable
3434
from . import SolutionLibrary
35-
from . import YAMLIO
35+
from . import LibraryIO
3636
from . import Utils
3737
from .BenchmarkStructs import BenchmarkProcess
3838
from .ClientWriter import runClient, writeClientParameters, writeClientConfig
@@ -362,7 +362,7 @@ def benchmarkProblemType( problemTypeConfig, problemSizeGroupConfig, \
362362
############################################################################
363363
# Write Solutions YAML
364364
############################################################################
365-
YAMLIO.writeSolutions(solutionsFileName, benchmarkStep.problemSizes, \
365+
LibraryIO.writeSolutions(solutionsFileName, benchmarkStep.problemSizes, \
366366
solutions )
367367

368368
# End Iteration
@@ -520,11 +520,12 @@ def writeBenchmarkFiles(stepBaseDir, solutions, problemSizes, stepName, filesToC
520520
globalParameters["WorkingPath"], globalParameters["CxxCompiler"], [problemType], solutions, kernels, kernelsBetaOnly, \
521521
solutionWriter, kernelWriterSource, kernelWriterAssembly, errorTolerant=True )
522522

523+
newLibraryFilename = "TensileLibrary.yaml" if globalParameters["YAML"] else "TensileLibrary.dat"
523524
newLibraryDir = ensurePath(os.path.join(globalParameters["WorkingPath"], 'library'))
524-
newLibraryFile = os.path.join(newLibraryDir, "TensileLibrary.yaml")
525+
newLibraryFile = os.path.join(newLibraryDir, newLibraryFilename)
525526
newLibrary = SolutionLibrary.MasterSolutionLibrary.BenchmarkingLibrary(solutions)
526527
newLibrary.applyNaming(kernelMinNaming)
527-
YAMLIO.write(newLibraryFile, Utils.state(newLibrary))
528+
LibraryIO.configWriter(globalParameters["YAML"]).write(newLibraryFile, Utils.state(newLibrary))
528529

529530
codeObjectFiles = [os.path.relpath(f, globalParameters["WorkingPath"]) for f in codeObjectFiles]
530531

Diff for: ‎Tensile/ClientWriter.py

+4-3
Original file line numberDiff line numberDiff line change
@@ -22,7 +22,7 @@
2222
from .Common import globalParameters, HR, pushWorkingPath, popWorkingPath, print1, CHeader, printWarning, listToInitializer, ClientExecutionLock
2323
from . import ClientExecutable
2424
from . import Common
25-
from . import YAMLIO
25+
from . import LibraryIO
2626

2727
import os
2828
import subprocess
@@ -98,7 +98,7 @@ def main( config ):
9898
for logicFileName in logicFiles:
9999
(scheduleName, deviceNames, problemType, solutionsForType, \
100100
indexOrder, exactLogic, rangeLogic, newLibrary, architectureName) \
101-
= YAMLIO.readLibraryLogicForSchedule(logicFileName)
101+
= LibraryIO.readLibraryLogicForSchedule(logicFileName)
102102
if problemType["DataType"].isHalf():
103103
enableHalf = True
104104
functions.append((scheduleName, problemType))
@@ -559,7 +559,8 @@ def param(key, value):
559559
f.write("{}={}\n".format(key, value))
560560

561561
sourceDir = os.path.join(stepBaseDir, "source")
562-
libraryFile = os.path.join(sourceDir, "library", "TensileLibrary.yaml")
562+
libraryFilename = "TensileLibrary.yaml" if globalParameters["YAML"] else "TensileLibrary.dat"
563+
libraryFile = os.path.join(sourceDir, "library", libraryFilename)
563564
param("library-file", libraryFile)
564565

565566
currentGFXName = Common.gfxName(globalParameters["CurrentISA"])

Diff for: ‎Tensile/Common.py

+1
Original file line numberDiff line numberDiff line change
@@ -108,6 +108,7 @@
108108
########################################
109109
globalParameters["CMakeBuildType"] = "Release" # whether benchmark clients and library client should be release or debug
110110
globalParameters["PrintSolutionRejectionReason"] = False # when a solution is marked as invalid, print why
111+
globalParameters["YAML"] = False # whether to use the YAML backend (as opposed to the MessagePack backend)
111112

112113
# how to initialize tensor data
113114
# serial-in-u will use a sequence that increments in the K dimension

Diff for: ‎Tensile/Component.py

+2-2
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,5 @@
11
################################################################################
2-
# Copyright (C) 2020 Advanced Micro Devices, Inc. All rights reserved.
2+
# Copyright 2020 Advanced Micro Devices, Inc. All rights reserved.
33
#
44
# Permission is hereby granted, free of charge, to any person obtaining a copy
55
# of this software and associated documentation files (the "Software"), to deal
@@ -72,7 +72,7 @@ def PartialMatch(pattern, obj):
7272
for key, value in pattern.items():
7373
if key not in obj:
7474
return False
75-
75+
7676
objValue = obj[key]
7777
if isinstance(value, collections.abc.Mapping) and \
7878
isinstance(objValue, collections.abc.Mapping):

Diff for: ‎Tensile/Components/__init__.py

+1-1
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
11

22
def use(): pass
33

4-
__all__ = ["MAC_F16"]
4+
__all__ = ["MAC_F16"]

Diff for: ‎Tensile/KernelWriterAssembly.py

+2-2
Original file line numberDiff line numberDiff line change
@@ -6027,7 +6027,7 @@ def endSummation(self, kernel):
60276027
# lastVgprForReads ^ ^ ^
60286028
# startVgprReuse ^ ^
60296029
# lastValuC ^
6030-
# if valuC does not include all of lastVgprForReads, we can reuse the
6030+
# if valuC does not include all of lastVgprForReads, we can reuse the
60316031
# non-overlapped part of lastVgprForReads
60326032
# |<-------------- valuC -------------->|
60336033
# |xxxxxxxxxxxxxxxxxxxxx|xxxxxxxxxxxxxxx|oooooo|xx|
@@ -6036,7 +6036,7 @@ def endSummation(self, kernel):
60366036
# startVgprReuse ^
60376037
vbegin = self.numVgprValuC
60386038
vsize = max(0, self.lastVgprForReads-self.numVgprValuC)
6039-
else:
6039+
else:
60406040
vbegin = self.startVgprValuA
60416041
vsize = self.lastVgprForReads - self.startVgprValuA
60426042

0 commit comments

Comments
 (0)