Skip to content

Commit f846e24

Browse files
yxsamliuyxsamliu
committed
[HIP] add --offload-compression-level= option
Added --offload-compression-level= option to clang and -compression-level= option to clang-offload-bundler for controlling compression level. Change default compression level to 20 for zstd for better compression rate.
1 parent ccc48d4 commit f846e24

File tree

12 files changed

+141
-46
lines changed

12 files changed

+141
-46
lines changed

‎clang/include/clang/Driver/OffloadBundler.h

Lines changed: 5 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -17,6 +17,7 @@
1717
#ifndef LLVM_CLANG_DRIVER_OFFLOADBUNDLER_H
1818
#define LLVM_CLANG_DRIVER_OFFLOADBUNDLER_H
1919

20+
#include "llvm/Support/Compression.h"
2021
#include "llvm/Support/Error.h"
2122
#include "llvm/TargetParser/Triple.h"
2223
#include <llvm/Support/MemoryBuffer.h>
@@ -36,6 +37,8 @@ class OffloadBundlerConfig {
3637
bool HipOpenmpCompatible = false;
3738
bool Compress = false;
3839
bool Verbose = false;
40+
llvm::compression::Format CompressionFormat;
41+
int CompressionLevel;
3942

4043
unsigned BundleAlignment = 1;
4144
unsigned HostInputIndex = ~0u;
@@ -116,7 +119,8 @@ class CompressedOffloadBundle {
116119

117120
public:
118121
static llvm::Expected<std::unique_ptr<llvm::MemoryBuffer>>
119-
compress(const llvm::MemoryBuffer &Input, bool Verbose = false);
122+
compress(llvm::compression::Params P, const llvm::MemoryBuffer &Input,
123+
bool Verbose = false);
120124
static llvm::Expected<std::unique_ptr<llvm::MemoryBuffer>>
121125
decompress(const llvm::MemoryBuffer &Input, bool Verbose = false);
122126
};

‎clang/include/clang/Driver/Options.td

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1264,6 +1264,10 @@ def fno_gpu_sanitize : Flag<["-"], "fno-gpu-sanitize">, Group<f_Group>;
12641264
def offload_compress : Flag<["--"], "offload-compress">,
12651265
HelpText<"Compress offload device binaries (HIP only)">;
12661266
def no_offload_compress : Flag<["--"], "no-offload-compress">;
1267+
1268+
def offload_compression_level_EQ : Joined<["--"], "offload-compression-level=">,
1269+
Flags<[HelpHidden]>,
1270+
HelpText<"Compression level for offload device binaries (HIP only)">;
12671271
}
12681272

12691273
// CUDA options

‎clang/lib/Driver/OffloadBundler.cpp

Lines changed: 65 additions & 25 deletions
Original file line numberDiff line numberDiff line change
@@ -906,6 +906,16 @@ CreateFileHandler(MemoryBuffer &FirstInput,
906906
}
907907

908908
OffloadBundlerConfig::OffloadBundlerConfig() {
909+
if (llvm::compression::zstd::isAvailable()) {
910+
CompressionFormat = llvm::compression::Format::Zstd;
911+
// Use a high zstd compress level by default for better size reduction.
912+
CompressionLevel = 20;
913+
} else if (llvm::compression::zlib::isAvailable()) {
914+
CompressionFormat = llvm::compression::Format::Zlib;
915+
// Use default level for zlib since higher level does not have significant
916+
// improvement.
917+
CompressionLevel = llvm::compression::zlib::DefaultCompression;
918+
}
909919
auto IgnoreEnvVarOpt =
910920
llvm::sys::Process::GetEnv("OFFLOAD_BUNDLER_IGNORE_ENV_VAR");
911921
if (IgnoreEnvVarOpt.has_value() && IgnoreEnvVarOpt.value() == "1")
@@ -919,11 +929,41 @@ OffloadBundlerConfig::OffloadBundlerConfig() {
919929
llvm::sys::Process::GetEnv("OFFLOAD_BUNDLER_COMPRESS");
920930
if (CompressEnvVarOpt.has_value())
921931
Compress = CompressEnvVarOpt.value() == "1";
932+
933+
auto CompressionLevelEnvVarOpt =
934+
llvm::sys::Process::GetEnv("OFFLOAD_BUNDLER_COMPRESSION_LEVEL");
935+
if (CompressionLevelEnvVarOpt.has_value()) {
936+
llvm::StringRef CompressionLevelStr = CompressionLevelEnvVarOpt.value();
937+
int Level;
938+
if (!CompressionLevelStr.getAsInteger(10, Level))
939+
CompressionLevel = Level;
940+
else
941+
llvm::errs()
942+
<< "Warning: Invalid value for OFFLOAD_BUNDLER_COMPRESSION_LEVEL: "
943+
<< CompressionLevelStr.str() << ". Ignoring it.\n";
944+
}
945+
}
946+
947+
// Utility function to format numbers with commas
948+
static std::string formatWithCommas(unsigned long long Value) {
949+
std::string Num = std::to_string(Value);
950+
int InsertPosition = Num.length() - 3;
951+
while (InsertPosition > 0) {
952+
Num.insert(InsertPosition, ",");
953+
InsertPosition -= 3;
954+
}
955+
return Num;
922956
}
923957

924958
llvm::Expected<std::unique_ptr<llvm::MemoryBuffer>>
925-
CompressedOffloadBundle::compress(const llvm::MemoryBuffer &Input,
959+
CompressedOffloadBundle::compress(llvm::compression::Params P,
960+
const llvm::MemoryBuffer &Input,
926961
bool Verbose) {
962+
if (!llvm::compression::zstd::isAvailable() &&
963+
!llvm::compression::zlib::isAvailable())
964+
return createStringError(llvm::inconvertibleErrorCode(),
965+
"Compression not supported");
966+
927967
llvm::Timer HashTimer("Hash Calculation Timer", "Hash calculation time",
928968
ClangOffloadBundlerTimerGroup);
929969
if (Verbose)
@@ -941,25 +981,15 @@ CompressedOffloadBundle::compress(const llvm::MemoryBuffer &Input,
941981
reinterpret_cast<const uint8_t *>(Input.getBuffer().data()),
942982
Input.getBuffer().size());
943983

944-
llvm::compression::Format CompressionFormat;
945-
946-
if (llvm::compression::zstd::isAvailable())
947-
CompressionFormat = llvm::compression::Format::Zstd;
948-
else if (llvm::compression::zlib::isAvailable())
949-
CompressionFormat = llvm::compression::Format::Zlib;
950-
else
951-
return createStringError(llvm::inconvertibleErrorCode(),
952-
"Compression not supported");
953-
954984
llvm::Timer CompressTimer("Compression Timer", "Compression time",
955985
ClangOffloadBundlerTimerGroup);
956986
if (Verbose)
957987
CompressTimer.startTimer();
958-
llvm::compression::compress(CompressionFormat, BufferUint8, CompressedBuffer);
988+
llvm::compression::compress(P, BufferUint8, CompressedBuffer);
959989
if (Verbose)
960990
CompressTimer.stopTimer();
961991

962-
uint16_t CompressionMethod = static_cast<uint16_t>(CompressionFormat);
992+
uint16_t CompressionMethod = static_cast<uint16_t>(P.format);
963993
uint32_t UncompressedSize = Input.getBuffer().size();
964994

965995
SmallVector<char, 0> FinalBuffer;
@@ -977,17 +1007,22 @@ CompressedOffloadBundle::compress(const llvm::MemoryBuffer &Input,
9771007

9781008
if (Verbose) {
9791009
auto MethodUsed =
980-
CompressionFormat == llvm::compression::Format::Zstd ? "zstd" : "zlib";
1010+
P.format == llvm::compression::Format::Zstd ? "zstd" : "zlib";
1011+
double CompressionRate =
1012+
static_cast<double>(UncompressedSize) / CompressedBuffer.size();
1013+
9811014
llvm::errs() << "Compressed bundle format version: " << Version << "\n"
9821015
<< "Compression method used: " << MethodUsed << "\n"
983-
<< "Binary size before compression: " << UncompressedSize
984-
<< " bytes\n"
985-
<< "Binary size after compression: " << CompressedBuffer.size()
986-
<< " bytes\n"
1016+
<< "Compression level: " << P.level << "\n"
1017+
<< "Binary size before compression: "
1018+
<< formatWithCommas(UncompressedSize) << " bytes\n"
1019+
<< "Binary size after compression: "
1020+
<< formatWithCommas(CompressedBuffer.size()) << " bytes\n"
1021+
<< "Compression rate: "
1022+
<< llvm::format("%.2lf", CompressionRate) << "\n"
9871023
<< "Truncated MD5 hash: "
9881024
<< llvm::format_hex(TruncatedHash, 16) << "\n";
9891025
}
990-
9911026
return llvm::MemoryBuffer::getMemBufferCopy(
9921027
llvm::StringRef(FinalBuffer.data(), FinalBuffer.size()));
9931028
}
@@ -1066,16 +1101,20 @@ CompressedOffloadBundle::decompress(const llvm::MemoryBuffer &Input,
10661101
HashRecalcTimer.stopTimer();
10671102
bool HashMatch = (StoredHash == RecalculatedHash);
10681103

1104+
double CompressionRate =
1105+
static_cast<double>(UncompressedSize) / CompressedData.size();
10691106
llvm::errs() << "Compressed bundle format version: " << ThisVersion << "\n"
10701107
<< "Decompression method: "
10711108
<< (CompressionFormat == llvm::compression::Format::Zlib
10721109
? "zlib"
10731110
: "zstd")
10741111
<< "\n"
1075-
<< "Size before decompression: " << CompressedData.size()
1076-
<< " bytes\n"
1077-
<< "Size after decompression: " << UncompressedSize
1078-
<< " bytes\n"
1112+
<< "Size before decompression: "
1113+
<< formatWithCommas(CompressedData.size()) << " bytes\n"
1114+
<< "Size after decompression: "
1115+
<< formatWithCommas(UncompressedSize) << " bytes\n"
1116+
<< "Compression rate: "
1117+
<< llvm::format("%.2lf", CompressionRate) << "\n"
10791118
<< "Stored hash: " << llvm::format_hex(StoredHash, 16) << "\n"
10801119
<< "Recalculated hash: "
10811120
<< llvm::format_hex(RecalculatedHash, 16) << "\n"
@@ -1269,8 +1308,9 @@ Error OffloadBundler::BundleFiles() {
12691308
std::unique_ptr<llvm::MemoryBuffer> BufferMemory =
12701309
llvm::MemoryBuffer::getMemBufferCopy(
12711310
llvm::StringRef(Buffer.data(), Buffer.size()));
1272-
auto CompressionResult =
1273-
CompressedOffloadBundle::compress(*BufferMemory, BundlerConfig.Verbose);
1311+
auto CompressionResult = CompressedOffloadBundle::compress(
1312+
{BundlerConfig.CompressionFormat, BundlerConfig.CompressionLevel},
1313+
*BufferMemory, BundlerConfig.Verbose);
12741314
if (auto Error = CompressionResult.takeError())
12751315
return Error;
12761316

‎clang/lib/Driver/ToolChains/Clang.cpp

Lines changed: 15 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -8518,6 +8518,20 @@ void ClangAs::ConstructJob(Compilation &C, const JobAction &JA,
85188518
}
85198519

85208520
// Begin OffloadBundler
8521+
void OffloadBundler::addCommonArgs(const llvm::opt::ArgList &TCArgs,
8522+
llvm::opt::ArgStringList &CmdArgs) {
8523+
if (TCArgs.hasFlag(options::OPT_offload_compress,
8524+
options::OPT_no_offload_compress, false))
8525+
CmdArgs.push_back("-compress");
8526+
if (TCArgs.hasArg(options::OPT_v))
8527+
CmdArgs.push_back("-verbose");
8528+
if (auto *Arg =
8529+
TCArgs.getLastArg(options::OPT_offload_compression_level_EQ)) {
8530+
std::string CompressionLevelArg =
8531+
std::string("-compression-level=") + Arg->getValue();
8532+
CmdArgs.push_back(TCArgs.MakeArgString(CompressionLevelArg));
8533+
}
8534+
}
85218535

85228536
void OffloadBundler::ConstructJob(Compilation &C, const JobAction &JA,
85238537
const InputInfo &Output,
@@ -8616,11 +8630,7 @@ void OffloadBundler::ConstructJob(Compilation &C, const JobAction &JA,
86168630
}
86178631
CmdArgs.push_back(TCArgs.MakeArgString(UB));
86188632
}
8619-
if (TCArgs.hasFlag(options::OPT_offload_compress,
8620-
options::OPT_no_offload_compress, false))
8621-
CmdArgs.push_back("-compress");
8622-
if (TCArgs.hasArg(options::OPT_v))
8623-
CmdArgs.push_back("-verbose");
8633+
addCommonArgs(TCArgs, CmdArgs);
86248634
// All the inputs are encoded as commands.
86258635
C.addCommand(std::make_unique<Command>(
86268636
JA, *this, ResponseFileSupport::None(),

‎clang/lib/Driver/ToolChains/Clang.h

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -157,6 +157,8 @@ class LLVM_LIBRARY_VISIBILITY OffloadBundler final : public Tool {
157157
const InputInfoList &Inputs,
158158
const llvm::opt::ArgList &TCArgs,
159159
const char *LinkingOutput) const override;
160+
static void addCommonArgs(const llvm::opt::ArgList &TCArgs,
161+
llvm::opt::ArgStringList &CmdArgs);
160162
};
161163

162164
/// Offload binary tool.

‎clang/lib/Driver/ToolChains/HIPUtility.cpp

Lines changed: 2 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -7,6 +7,7 @@
77
//===----------------------------------------------------------------------===//
88

99
#include "HIPUtility.h"
10+
#include "Clang.h"
1011
#include "CommonArgs.h"
1112
#include "clang/Driver/Compilation.h"
1213
#include "clang/Driver/Options.h"
@@ -258,11 +259,7 @@ void HIP::constructHIPFatbinCommand(Compilation &C, const JobAction &JA,
258259
Args.MakeArgString(std::string("-output=").append(Output));
259260
BundlerArgs.push_back(BundlerOutputArg);
260261

261-
if (Args.hasFlag(options::OPT_offload_compress,
262-
options::OPT_no_offload_compress, false))
263-
BundlerArgs.push_back("-compress");
264-
if (Args.hasArg(options::OPT_v))
265-
BundlerArgs.push_back("-verbose");
262+
OffloadBundler::addCommonArgs(Args, BundlerArgs);
266263

267264
const char *Bundler = Args.MakeArgString(
268265
T.getToolChain().GetProgramPath("clang-offload-bundler"));

‎clang/test/Driver/clang-offload-bundler-zlib.c

Lines changed: 18 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
// REQUIRES: zlib
1+
// REQUIRES: zlib && !zstd
22
// REQUIRES: x86-registered-target
33
// UNSUPPORTED: target={{.*}}-darwin{{.*}}, target={{.*}}-aix{{.*}}
44

@@ -34,13 +34,28 @@
3434
// RUN: diff %t.tgt2 %t.res.tgt2
3535

3636
//
37-
// COMPRESS: Compression method used:
38-
// DECOMPRESS: Decompression method:
37+
// COMPRESS: Compression method used: zlib
38+
// COMPRESS: Compression level: 6
39+
// DECOMPRESS: Decompression method: zlib
40+
// DECOMPRESS: Hashes match: Yes
3941
// NOHOST-NOT: host-
4042
// NOHOST-DAG: hip-amdgcn-amd-amdhsa--gfx900
4143
// NOHOST-DAG: hip-amdgcn-amd-amdhsa--gfx906
4244
//
4345

46+
// Check -compression-level= option
47+
48+
// RUN: clang-offload-bundler -type=bc -targets=hip-amdgcn-amd-amdhsa--gfx900,hip-amdgcn-amd-amdhsa--gfx906 \
49+
// RUN: -input=%t.tgt1 -input=%t.tgt2 -output=%t.hip.bundle.bc -compress -verbose -compression-level=9 2>&1 | \
50+
// RUN: FileCheck -check-prefix=LEVEL %s
51+
// RUN: clang-offload-bundler -type=bc -targets=hip-amdgcn-amd-amdhsa--gfx900,hip-amdgcn-amd-amdhsa--gfx906 \
52+
// RUN: -output=%t.res.tgt1 -output=%t.res.tgt2 -input=%t.hip.bundle.bc -unbundle
53+
// RUN: diff %t.tgt1 %t.res.tgt1
54+
// RUN: diff %t.tgt2 %t.res.tgt2
55+
//
56+
// LEVEL: Compression method used: zlib
57+
// LEVEL: Compression level: 9
58+
4459
//
4560
// Check -bundle-align option.
4661
//

‎clang/test/Driver/clang-offload-bundler-zstd.c

Lines changed: 17 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -31,13 +31,28 @@
3131
// RUN: diff %t.tgt1 %t.res.tgt1
3232
// RUN: diff %t.tgt2 %t.res.tgt2
3333
//
34-
// COMPRESS: Compression method used
35-
// DECOMPRESS: Decompression method
34+
// COMPRESS: Compression method used: zstd
35+
// COMPRESS: Compression level: 20
36+
// DECOMPRESS: Decompression method: zstd
37+
// DECOMPRESS: Hashes match: Yes
3638
// NOHOST-NOT: host-
3739
// NOHOST-DAG: hip-amdgcn-amd-amdhsa--gfx900
3840
// NOHOST-DAG: hip-amdgcn-amd-amdhsa--gfx906
3941
//
4042

43+
// Check -compression-level= option
44+
45+
// RUN: clang-offload-bundler -type=bc -targets=hip-amdgcn-amd-amdhsa--gfx900,hip-amdgcn-amd-amdhsa--gfx906 \
46+
// RUN: -input=%t.tgt1 -input=%t.tgt2 -output=%t.hip.bundle.bc -compress -verbose -compression-level=9 2>&1 | \
47+
// RUN: FileCheck -check-prefix=LEVEL %s
48+
// RUN: clang-offload-bundler -type=bc -targets=hip-amdgcn-amd-amdhsa--gfx900,hip-amdgcn-amd-amdhsa--gfx906 \
49+
// RUN: -output=%t.res.tgt1 -output=%t.res.tgt2 -input=%t.hip.bundle.bc -unbundle
50+
// RUN: diff %t.tgt1 %t.res.tgt1
51+
// RUN: diff %t.tgt2 %t.res.tgt2
52+
//
53+
// LEVEL: Compression method used: zstd
54+
// LEVEL: Compression level: 9
55+
4156
//
4257
// Check -bundle-align option.
4358
//

‎clang/test/Driver/hip-offload-compress-zlib.hip

Lines changed: 4 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
// REQUIRES: zlib
1+
// REQUIRES: zlib && !zstd
22
// REQUIRES: x86-registered-target
33
// REQUIRES: amdgpu-registered-target
44

@@ -9,13 +9,14 @@
99
// RUN: -x hip --offload-arch=gfx1100 --offload-arch=gfx1101 \
1010
// RUN: -fgpu-rdc -nogpuinc -nogpulib \
1111
// RUN: %S/Inputs/hip_multiple_inputs/a.cu \
12-
// RUN: --offload-compress --offload-device-only --gpu-bundle-output \
12+
// RUN: --offload-compress --offload-compression-level=9 \
13+
// RUN: --offload-device-only --gpu-bundle-output \
1314
// RUN: -o %t.bc \
1415
// RUN: 2>&1 | FileCheck %s
1516

1617
// CHECK: clang-offload-bundler{{.*}} -type=bc
1718
// CHECK-SAME: -targets={{.*}}hip-amdgcn-amd-amdhsa-gfx1100,hip-amdgcn-amd-amdhsa-gfx1101
18-
// CHECK-SAME: -compress -verbose
19+
// CHECK-SAME: -compress -verbose -compression-level=9
1920
// CHECK: Compressed bundle format
2021

2122
// Test uncompress of bundled bitcode.

‎clang/test/Driver/hip-offload-compress-zstd.hip

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -9,13 +9,14 @@
99
// RUN: -x hip --offload-arch=gfx1100 --offload-arch=gfx1101 \
1010
// RUN: -fgpu-rdc -nogpuinc -nogpulib \
1111
// RUN: %S/Inputs/hip_multiple_inputs/a.cu \
12-
// RUN: --offload-compress --offload-device-only --gpu-bundle-output \
12+
// RUN: --offload-compress --offload-compression-level=9 \
13+
// RUN: --offload-device-only --gpu-bundle-output \
1314
// RUN: -o %t.bc \
1415
// RUN: 2>&1 | FileCheck %s
1516

1617
// CHECK: clang-offload-bundler{{.*}} -type=bc
1718
// CHECK-SAME: -targets={{.*}}hip-amdgcn-amd-amdhsa-gfx1100,hip-amdgcn-amd-amdhsa-gfx1101
18-
// CHECK-SAME: -compress -verbose
19+
// CHECK-SAME: -compress -verbose -compression-level=9
1920
// CHECK: Compressed bundle format
2021

2122
// Test uncompress of bundled bitcode.

‎clang/tools/clang-offload-bundler/ClangOffloadBundler.cpp

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -145,6 +145,9 @@ int main(int argc, const char **argv) {
145145
cl::init(false), cl::cat(ClangOffloadBundlerCategory));
146146
cl::opt<bool> Verbose("verbose", cl::desc("Print debug information.\n"),
147147
cl::init(false), cl::cat(ClangOffloadBundlerCategory));
148+
cl::opt<int> CompressionLevel(
149+
"compression-level", cl::desc("Specify the compression level (integer)"),
150+
cl::value_desc("n"), cl::Optional, cl::cat(ClangOffloadBundlerCategory));
148151

149152
// Process commandline options and report errors
150153
sys::PrintStackTraceOnErrorSignal(argv[0]);
@@ -178,6 +181,8 @@ int main(int argc, const char **argv) {
178181
BundlerConfig.Compress = Compress;
179182
if (Verbose.getNumOccurrences() > 0)
180183
BundlerConfig.Verbose = Verbose;
184+
if (CompressionLevel.getNumOccurrences() > 0)
185+
BundlerConfig.CompressionLevel = CompressionLevel;
181186

182187
BundlerConfig.TargetNames = TargetNames;
183188
BundlerConfig.InputFileNames = InputFileNames;

‎llvm/include/llvm/Support/Compression.h

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -94,6 +94,7 @@ struct Params {
9494
constexpr Params(Format F)
9595
: format(F), level(F == Format::Zlib ? zlib::DefaultCompression
9696
: zstd::DefaultCompression) {}
97+
constexpr Params(Format F, int L) : format(F), level(L) {}
9798
Params(DebugCompressionType Type) : Params(formatFor(Type)) {}
9899

99100
Format format;

0 commit comments

Comments
 (0)