[libc] NVPTX Profiling #92009

jameshu15869 · 2024-05-13T18:46:41Z

PR for adding microbenchmarking infrastructure for NVPTX. nvlink cannot perform LTO, so we cannot inline libc functions and this function call overhead is not adjusted for during microbenchmarking.

llvmbot · 2024-05-13T18:47:11Z

@llvm/pr-subscribers-libc

Author: None (jameshu15869)

Changes

Draft PR for adding microbenchmarking infrastructure for NVPTX. nvlink cannot perform LTO, so we cannot inline libc functions and this function call overhead is not adjusted for during microbenchmarking.

Patch is 38.98 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/92009.diff

16 Files Affected:

(modified) libc/benchmarks/CMakeLists.txt (+211-205)
(added) libc/benchmarks/gpu/BenchmarkLogger.cpp (+89)
(added) libc/benchmarks/gpu/BenchmarkLogger.h (+27)
(added) libc/benchmarks/gpu/CMakeLists.txt (+183)
(added) libc/benchmarks/gpu/LibcGpuBenchmark.cpp (+70)
(added) libc/benchmarks/gpu/LibcGpuBenchmark.h (+122)
(added) libc/benchmarks/gpu/LibcGpuBenchmarkMain.cpp (+6)
(added) libc/benchmarks/gpu/src/CMakeLists.txt (+2)
(added) libc/benchmarks/gpu/src/ctype/CMakeLists.txt (+21)
(added) libc/benchmarks/gpu/src/ctype/isalnum_benchmark.cpp (+22)
(added) libc/benchmarks/gpu/src/ctype/isalpha_benchmark.cpp (+9)
(added) libc/benchmarks/gpu/src/math/CMakeLists.txt ()
(added) libc/benchmarks/gpu/timing/CMakeLists.txt (+12)
(added) libc/benchmarks/gpu/timing/nvptx/CMakeLists.txt (+7)
(added) libc/benchmarks/gpu/timing/nvptx/timing.h (+108)
(added) libc/benchmarks/gpu/timing/timing.h (+23)

diff --git a/libc/benchmarks/CMakeLists.txt b/libc/benchmarks/CMakeLists.txt
index 4978da65850cc..a802e653a091e 100644
--- a/libc/benchmarks/CMakeLists.txt
+++ b/libc/benchmarks/CMakeLists.txt
@@ -1,205 +1,211 @@
-find_package(Threads)
-
-set(LLVM_LINK_COMPONENTS
-  Support
-  TargetParser
-  )
-
-#==============================================================================
-# Add Unit Testing Support
-#==============================================================================
-
-function(add_libc_benchmark_unittest target_name)
-  if(NOT LLVM_INCLUDE_TESTS)
-    return()
-  endif()
-
-  cmake_parse_arguments(
-    "LIBC_BENCHMARKS_UNITTEST"
-    "" # No optional arguments
-    "SUITE" # Single value arguments
-    "SRCS;DEPENDS" # Multi-value arguments
-    ${ARGN}
-  )
-
-  add_executable(${target_name}
-    EXCLUDE_FROM_ALL
-    ${LIBC_BENCHMARKS_UNITTEST_SRCS}
-  )
-  target_link_libraries(${target_name}
-    PRIVATE
-    llvm_gtest_main
-    llvm_gtest
-    ${LIBC_BENCHMARKS_UNITTEST_DEPENDS}
-  )
-  llvm_update_compile_flags(${target_name})
-
-  add_custom_command(
-    TARGET ${target_name}
-    POST_BUILD
-    COMMAND $<TARGET_FILE:${target_name}>
-  )
-  add_dependencies(libc-benchmark-util-tests ${target_name})
-endfunction()
-
-#==============================================================================
-# Build Google Benchmark for libc
-#==============================================================================
-
-include(ExternalProject)
-ExternalProject_Add(google-benchmark-libc
-        EXCLUDE_FROM_ALL ON
-        PREFIX google-benchmark-libc
-        SOURCE_DIR ${LLVM_THIRD_PARTY_DIR}/benchmark
-        INSTALL_DIR ${CMAKE_CURRENT_BINARY_DIR}/google-benchmark-libc
-        CMAKE_CACHE_ARGS
-          -DBENCHMARK_ENABLE_EXCEPTIONS:BOOL=OFF
-          -DBENCHMARK_ENABLE_LTO:BOOL=OFF
-          -DBENCHMARK_ENABLE_TESTING:BOOL=OFF
-          -DBENCHMARK_ENABLE_WERROR:BOOL=${LLVM_ENABLE_WERROR}
-          -DBENCHMARK_FORCE_WERROR:BOOL=OFF
-          -DBENCHMARK_USE_LIBCXX:BOOL=OFF
-          -DCMAKE_BUILD_TYPE:STRING=Release
-
-          -DCMAKE_SYSTEM_NAME:STRING=${CMAKE_SYSTEM_NAME}
-          -DCMAKE_SYSTEM_PROCESSOR:STRING=${CMAKE_SYSTEM_PROCESSOR}
-          -DCMAKE_C_COMPILER:STRING=${CMAKE_C_COMPILER}
-          -DCMAKE_CXX_COMPILER:STRING=${CMAKE_CXX_COMPILER}
-          -DCMAKE_CXX_FLAGS:STRING=${CMAKE_CXX_FLAGS}
-          -DCMAKE_FIND_ROOT_PATH:STRING=${CMAKE_FIND_ROOT_PATH}
-
-          -DBUILD_SHARED_LIBS:BOOL=OFF
-          -DCMAKE_EXE_LINKER_FLAGS:STRING=-static
-
-          -DCMAKE_CXX_STANDARD:STRING=14
-          -DCMAKE_INSTALL_PREFIX:PATH=<INSTALL_DIR>
-        )
-
-add_custom_target(libc-benchmark-util-tests)
-
-# libc-benchmark
-add_library(libc-benchmark
-    STATIC
-    EXCLUDE_FROM_ALL
-    LibcBenchmark.cpp
-    LibcBenchmark.h
-)
-
-target_include_directories(libc-benchmark
-    PUBLIC ${LLVM_INCLUDE_DIR} ${LLVM_MAIN_INCLUDE_DIR}
-)
-target_link_libraries(libc-benchmark
-    PUBLIC
-    benchmark::benchmark
-    LLVMSupport
-    LLVMTargetParser
-    Threads::Threads
-)
-add_dependencies(libc-benchmark google-benchmark-libc)
-llvm_update_compile_flags(libc-benchmark)
-
-add_libc_benchmark_unittest(libc-benchmark-test
-    SRCS LibcBenchmarkTest.cpp
-    DEPENDS libc-benchmark
-)
-
-# libc-memory-benchmark
-add_library(libc-memory-benchmark
-    STATIC
-    EXCLUDE_FROM_ALL
-    LibcMemoryBenchmark.cpp
-    LibcMemoryBenchmark.h
-    LibcFunctionPrototypes.h
-    MemorySizeDistributions.cpp
-    MemorySizeDistributions.h
-)
-target_include_directories(libc-memory-benchmark
-    PUBLIC
-    ${CMAKE_CURRENT_SOURCE_DIR}
-)
-target_link_libraries(libc-memory-benchmark
-    PUBLIC
-    libc-benchmark
-)
-llvm_update_compile_flags(libc-memory-benchmark)
-
-add_libc_benchmark_unittest(libc-memory-benchmark-test
-    SRCS LibcMemoryBenchmarkTest.cpp
-    DEPENDS libc-memory-benchmark
-)
-
-# json
-add_library(json
-    STATIC
-    EXCLUDE_FROM_ALL
-    JSON.cpp
-    JSON.h
-)
-target_link_libraries(json PUBLIC libc-memory-benchmark)
-llvm_update_compile_flags(json)
-
-add_libc_benchmark_unittest(json-test
-    SRCS JSONTest.cpp
-    DEPENDS json
-)
-
-#==============================================================================
-# Benchmarking tool
-#==============================================================================
-
-# Benchmark all implementations that can run on the target CPU.
-function(add_libc_multi_impl_benchmark name)
-  get_property(fq_implementations GLOBAL PROPERTY ${name}_implementations)
-  foreach(fq_config_name IN LISTS fq_implementations)
-    get_target_property(required_cpu_features ${fq_config_name} REQUIRE_CPU_FEATURES)
-    cpu_supports(can_run "${required_cpu_features}")
-    if(can_run)
-        set(benchmark_name ${fq_config_name}_benchmark)
-        add_executable(${benchmark_name}
-            EXCLUDE_FROM_ALL
-            LibcMemoryBenchmarkMain.cpp
-        )
-        get_target_property(entrypoint_object_file ${fq_config_name} "OBJECT_FILE_RAW")
-        target_link_libraries(${benchmark_name} PUBLIC json ${entrypoint_object_file})
-        string(TOUPPER ${name} name_upper)
-        target_compile_definitions(${benchmark_name} PRIVATE "-DLIBC_BENCHMARK_FUNCTION_${name_upper}=LIBC_NAMESPACE::${name}" "-DLIBC_BENCHMARK_FUNCTION_NAME=\"${fq_config_name}\"")
-        llvm_update_compile_flags(${benchmark_name})
-    else()
-      message(STATUS "Skipping benchmark for '${fq_config_name}' insufficient host cpu features '${required_cpu_features}'")
-    endif()
-  endforeach()
-endfunction()
-
-add_libc_multi_impl_benchmark(bcmp)
-add_libc_multi_impl_benchmark(bzero)
-add_libc_multi_impl_benchmark(memcmp)
-add_libc_multi_impl_benchmark(memcpy)
-add_libc_multi_impl_benchmark(memmove)
-add_libc_multi_impl_benchmark(memset)
-
-#==============================================================================
-# Google Benchmarking tool
-#==============================================================================
-
-# This target uses the Google Benchmark facility to report throughput for llvm
-# libc memory functions compiled for the host machine. This is useful to
-# continuously monitor the performance of the memory functions.
-add_executable(libc.benchmarks.memory_functions.opt_host
-  EXCLUDE_FROM_ALL
-  LibcMemoryGoogleBenchmarkMain.cpp
-  LibcDefaultImplementations.cpp
-)
-target_link_libraries(libc.benchmarks.memory_functions.opt_host
-  PRIVATE
-  libc-memory-benchmark
-  libc.src.string.memcmp_opt_host.__internal__
-  libc.src.string.bcmp_opt_host.__internal__
-  libc.src.string.memcpy_opt_host.__internal__
-  libc.src.string.memset_opt_host.__internal__
-  libc.src.string.bzero_opt_host.__internal__
-  libc.src.string.memmove_opt_host.__internal__
-  benchmark_main
-)
-llvm_update_compile_flags(libc.benchmarks.memory_functions.opt_host)
-
-add_subdirectory(automemcpy)
+if(NOT LIBC_TARGET_OS_IS_GPU)
+	find_package(Threads)
+
+	set(LLVM_LINK_COMPONENTS
+	  Support
+	  TargetParser
+	  )
+
+	#==============================================================================
+	# Add Unit Testing Support
+	#==============================================================================
+
+	function(add_libc_benchmark_unittest target_name)
+	  if(NOT LLVM_INCLUDE_TESTS)
+	    return()
+	  endif()
+
+	  cmake_parse_arguments(
+	    "LIBC_BENCHMARKS_UNITTEST"
+	    "" # No optional arguments
+	    "SUITE" # Single value arguments
+	    "SRCS;DEPENDS" # Multi-value arguments
+	    ${ARGN}
+	  )
+
+	  add_executable(${target_name}
+	    EXCLUDE_FROM_ALL
+	    ${LIBC_BENCHMARKS_UNITTEST_SRCS}
+	  )
+	  target_link_libraries(${target_name}
+	    PRIVATE
+	    llvm_gtest_main
+	    llvm_gtest
+	    ${LIBC_BENCHMARKS_UNITTEST_DEPENDS}
+	  )
+	  llvm_update_compile_flags(${target_name})
+
+	  add_custom_command(
+	    TARGET ${target_name}
+	    POST_BUILD
+	    COMMAND $<TARGET_FILE:${target_name}>
+	  )
+	  add_dependencies(libc-benchmark-util-tests ${target_name})
+	endfunction()
+
+	#==============================================================================
+	# Build Google Benchmark for libc
+	#==============================================================================
+
+	include(ExternalProject)
+	ExternalProject_Add(google-benchmark-libc
+		EXCLUDE_FROM_ALL ON
+		PREFIX google-benchmark-libc
+		SOURCE_DIR ${LLVM_THIRD_PARTY_DIR}/benchmark
+		INSTALL_DIR ${CMAKE_CURRENT_BINARY_DIR}/google-benchmark-libc
+		CMAKE_CACHE_ARGS
+		  -DBENCHMARK_ENABLE_EXCEPTIONS:BOOL=OFF
+		  -DBENCHMARK_ENABLE_LTO:BOOL=OFF
+		  -DBENCHMARK_ENABLE_TESTING:BOOL=OFF
+		  -DBENCHMARK_ENABLE_WERROR:BOOL=${LLVM_ENABLE_WERROR}
+		  -DBENCHMARK_FORCE_WERROR:BOOL=OFF
+		  -DBENCHMARK_USE_LIBCXX:BOOL=OFF
+		  -DCMAKE_BUILD_TYPE:STRING=Release
+
+		  -DCMAKE_SYSTEM_NAME:STRING=${CMAKE_SYSTEM_NAME}
+		  -DCMAKE_SYSTEM_PROCESSOR:STRING=${CMAKE_SYSTEM_PROCESSOR}
+		  -DCMAKE_C_COMPILER:STRING=${CMAKE_C_COMPILER}
+		  -DCMAKE_CXX_COMPILER:STRING=${CMAKE_CXX_COMPILER}
+		  -DCMAKE_CXX_FLAGS:STRING=${CMAKE_CXX_FLAGS}
+		  -DCMAKE_FIND_ROOT_PATH:STRING=${CMAKE_FIND_ROOT_PATH}
+
+		  -DBUILD_SHARED_LIBS:BOOL=OFF
+		  -DCMAKE_EXE_LINKER_FLAGS:STRING=-static
+
+		  -DCMAKE_CXX_STANDARD:STRING=14
+		  -DCMAKE_INSTALL_PREFIX:PATH=<INSTALL_DIR>
+		)
+
+	add_custom_target(libc-benchmark-util-tests)
+
+	# libc-benchmark
+	add_library(libc-benchmark
+	    STATIC
+	    EXCLUDE_FROM_ALL
+	    LibcBenchmark.cpp
+	    LibcBenchmark.h
+	)
+
+	target_include_directories(libc-benchmark
+	    PUBLIC ${LLVM_INCLUDE_DIR} ${LLVM_MAIN_INCLUDE_DIR}
+	)
+	target_link_libraries(libc-benchmark
+	    PUBLIC
+	    benchmark::benchmark
+            LLVMSupport
+	    LLVMTargetParser
+	    Threads::Threads
+	)
+	add_dependencies(libc-benchmark google-benchmark-libc)
+	llvm_update_compile_flags(libc-benchmark)
+
+	add_libc_benchmark_unittest(libc-benchmark-test
+	    SRCS LibcBenchmarkTest.cpp
+	    DEPENDS libc-benchmark
+	)
+
+	# libc-memory-benchmark
+	add_library(libc-memory-benchmark
+	    STATIC
+	    EXCLUDE_FROM_ALL
+	    LibcMemoryBenchmark.cpp
+	    LibcMemoryBenchmark.h
+	    LibcFunctionPrototypes.h
+	    MemorySizeDistributions.cpp
+	    MemorySizeDistributions.h
+	)
+	target_include_directories(libc-memory-benchmark
+	    PUBLIC
+	    ${CMAKE_CURRENT_SOURCE_DIR}
+	)
+	target_link_libraries(libc-memory-benchmark
+	    PUBLIC
+	    libc-benchmark
+	)
+	llvm_update_compile_flags(libc-memory-benchmark)
+
+	add_libc_benchmark_unittest(libc-memory-benchmark-test
+	    SRCS LibcMemoryBenchmarkTest.cpp
+	    DEPENDS libc-memory-benchmark
+	)
+
+	# json
+	add_library(json
+	    STATIC
+	    EXCLUDE_FROM_ALL
+	    JSON.cpp
+	    JSON.h
+	)
+	target_link_libraries(json PUBLIC libc-memory-benchmark)
+	llvm_update_compile_flags(json)
+
+	add_libc_benchmark_unittest(json-test
+	    SRCS JSONTest.cpp
+	    DEPENDS json
+	)
+
+	#==============================================================================
+	# Benchmarking tool
+	#==============================================================================
+
+	# Benchmark all implementations that can run on the target CPU.
+	function(add_libc_multi_impl_benchmark name)
+	  get_property(fq_implementations GLOBAL PROPERTY ${name}_implementations)
+	  foreach(fq_config_name IN LISTS fq_implementations)
+	    get_target_property(required_cpu_features ${fq_config_name} REQUIRE_CPU_FEATURES)
+	    cpu_supports(can_run "${required_cpu_features}")
+	    if(can_run)
+		set(benchmark_name ${fq_config_name}_benchmark)
+		add_executable(${benchmark_name}
+		    EXCLUDE_FROM_ALL
+		    LibcMemoryBenchmarkMain.cpp
+		)
+		get_target_property(entrypoint_object_file ${fq_config_name} "OBJECT_FILE_RAW")
+		target_link_libraries(${benchmark_name} PUBLIC json ${entrypoint_object_file})
+		string(TOUPPER ${name} name_upper)
+		target_compile_definitions(${benchmark_name} PRIVATE "-DLIBC_BENCHMARK_FUNCTION_${name_upper}=LIBC_NAMESPACE::${name}" "-DLIBC_BENCHMARK_FUNCTION_NAME=\"${fq_config_name}\"")
+		llvm_update_compile_flags(${benchmark_name})
+	    else()
+	      message(STATUS "Skipping benchmark for '${fq_config_name}' insufficient host cpu features '${required_cpu_features}'")
+	    endif()
+	  endforeach()
+	endfunction()
+
+	add_libc_multi_impl_benchmark(bcmp)
+	add_libc_multi_impl_benchmark(bzero)
+	add_libc_multi_impl_benchmark(memcmp)
+	add_libc_multi_impl_benchmark(memcpy)
+	add_libc_multi_impl_benchmark(memmove)
+	add_libc_multi_impl_benchmark(memset)
+
+	#==============================================================================
+	# Google Benchmarking tool
+	#==============================================================================
+
+	# This target uses the Google Benchmark facility to report throughput for llvm
+	# libc memory functions compiled for the host machine. This is useful to
+	# continuously monitor the performance of the memory functions.
+	add_executable(libc.benchmarks.memory_functions.opt_host
+	  EXCLUDE_FROM_ALL
+	  LibcMemoryGoogleBenchmarkMain.cpp
+	  LibcDefaultImplementations.cpp
+	)
+	target_link_libraries(libc.benchmarks.memory_functions.opt_host
+	  PRIVATE
+	  libc-memory-benchmark
+	  libc.src.string.memcmp_opt_host.__internal__
+	  libc.src.string.bcmp_opt_host.__internal__
+	  libc.src.string.memcpy_opt_host.__internal__
+	  libc.src.string.memset_opt_host.__internal__
+	  libc.src.string.bzero_opt_host.__internal__
+	  libc.src.string.memmove_opt_host.__internal__
+	  benchmark_main
+	)
+	llvm_update_compile_flags(libc.benchmarks.memory_functions.opt_host)
+
+	add_subdirectory(automemcpy)
+endif()
+
+if(LIBC_TARGET_OS_IS_GPU)
+	add_subdirectory(gpu)
+endif()
diff --git a/libc/benchmarks/gpu/BenchmarkLogger.cpp b/libc/benchmarks/gpu/BenchmarkLogger.cpp
new file mode 100644
index 0000000000000..94a0d897c9585
--- /dev/null
+++ b/libc/benchmarks/gpu/BenchmarkLogger.cpp
@@ -0,0 +1,89 @@
+#include "benchmarks/gpu/BenchmarkLogger.h"
+#include "src/__support/CPP/string.h"
+#include "src/__support/CPP/string_view.h"
+#include "src/__support/OSUtil/io.h"               // write_to_stderr
+#include "src/__support/big_int.h"                 // is_big_int
+#include "src/__support/macros/properties/types.h" // LIBC_TYPES_HAS_INT128
+#include "src/__support/uint128.h"
+
+#include <stdint.h>
+
+namespace LIBC_NAMESPACE {
+namespace libc_gpu_benchmarks {
+
+// cpp::string_view specialization
+template <>
+BenchmarkLogger &BenchmarkLogger::operator<< <cpp::string_view>(cpp::string_view str) {
+  LIBC_NAMESPACE::write_to_stderr(str);
+  return *this;
+}
+
+// cpp::string specialization
+template <> BenchmarkLogger &BenchmarkLogger::operator<< <cpp::string>(cpp::string str) {
+  return *this << static_cast<cpp::string_view>(str);
+}
+
+// const char* specialization
+template <> BenchmarkLogger &BenchmarkLogger::operator<< <const char *>(const char *str) {
+  return *this << cpp::string_view(str);
+}
+
+// char* specialization
+template <> BenchmarkLogger &BenchmarkLogger::operator<< <char *>(char *str) {
+  return *this << cpp::string_view(str);
+}
+
+// char specialization
+template <> BenchmarkLogger &BenchmarkLogger::operator<<(char ch) {
+  return *this << cpp::string_view(&ch, 1);
+}
+
+// bool specialization
+template <> BenchmarkLogger &BenchmarkLogger::operator<<(bool cond) {
+  return *this << (cond ? "true" : "false");
+}
+
+// void * specialization
+template <> BenchmarkLogger &BenchmarkLogger::operator<<(void *addr) {
+  return *this << "0x" << cpp::to_string(reinterpret_cast<uintptr_t>(addr));
+}
+
+template <typename T> BenchmarkLogger &BenchmarkLogger::operator<<(T t) {
+  if constexpr (is_big_int_v<T> ||
+                (cpp::is_integral_v<T> && cpp::is_unsigned_v<T> &&
+                 (sizeof(T) > sizeof(uint64_t)))) {
+    static_assert(sizeof(T) % 8 == 0, "Unsupported size of UInt");
+    const IntegerToString<T, radix::Hex::WithPrefix> buffer(t);
+    return *this << buffer.view();
+  } else {
+    return *this << cpp::to_string(t);
+  }
+}
+
+// is_integral specializations
+// char is already specialized to handle character
+template BenchmarkLogger &BenchmarkLogger::operator<< <short>(short);
+template BenchmarkLogger &BenchmarkLogger::operator<< <int>(int);
+template BenchmarkLogger &BenchmarkLogger::operator<< <long>(long);
+template BenchmarkLogger &BenchmarkLogger::operator<< <long long>(long long);
+template BenchmarkLogger &BenchmarkLogger::operator<< <unsigned char>(unsigned char);
+template BenchmarkLogger &BenchmarkLogger::operator<< <unsigned short>(unsigned short);
+template BenchmarkLogger &BenchmarkLogger::operator<< <unsigned int>(unsigned int);
+template BenchmarkLogger &BenchmarkLogger::operator<< <unsigned long>(unsigned long);
+template BenchmarkLogger &
+    BenchmarkLogger::operator<< <unsigned long long>(unsigned long long);
+
+#ifdef LIBC_TYPES_HAS_INT128
+template BenchmarkLogger &BenchmarkLogger::operator<< <__uint128_t>(__uint128_t);
+#endif // LIBC_TYPES_HAS_INT128
+template BenchmarkLogger &BenchmarkLogger::operator<< <UInt<128>>(UInt<128>);
+template BenchmarkLogger &BenchmarkLogger::operator<< <UInt<192>>(UInt<192>);
+template BenchmarkLogger &BenchmarkLogger::operator<< <UInt<256>>(UInt<256>);
+template BenchmarkLogger &BenchmarkLogger::operator<< <UInt<320>>(UInt<320>);
+
+// TODO: Add floating point formatting once it's supported by StringStream.
+
+BenchmarkLogger blog;
+
+} // namespace libc_gpu_benchmarks
+} // namespace LIBC_NAMESPACE
diff --git a/libc/benchmarks/gpu/BenchmarkLogger.h b/libc/benchmarks/gpu/BenchmarkLogger.h
new file mode 100644
index 0000000000000..ed3cc97e59c6d
--- /dev/null
+++ b/libc/benchmarks/gpu/BenchmarkLogger.h
@@ -0,0 +1,27 @@
+//===-- Utilities to log to standard output during tests --------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_LIBC_BENCHMARKS_GPU_BENCHMARKLOGGER_H
+#define LLVM_LIBC_BENCHMARKS_GPU_BENCHMARKLOGGER_H
+
+namespace LIBC_NAMESPACE {
+namespace libc_gpu_benchmarks {
+
+// A class to log to standard output in the context of hermetic tests.
+struct BenchmarkLogger {
+  constexpr BenchmarkLogger() = default;
+  template <typename T> BenchmarkLogger &operator<<(T);
+};
+
+// A global TestLogger instance to be used in tests.
+extern BenchmarkLogger blog;
+
+} // namespace libc_gpu_benchmarks
+} // namespace LIBC_NAMESPACE
+
+#endif /* LLVM_LIBC_BENCHMARKS_GPU_BENCHMARKLOGGER_H */
diff --git a/libc/benchmarks/gpu/CMakeLists.txt b/libc/benchmarks/gpu/CMakeLists.txt
new file mode 100644
index 0000000000000..a18be27e33573
--- /dev/null
+++ b/libc/benchmarks/gpu/CMakeLists.txt
@@ -0,0 +1,183 @@
+add_subdirectory(timing)
+
+add_custom_target(gpu-benchmark)
+
+function (add_gpu_benchmark test_name)
+  if(NOT TARGET libc.startup.${LIBC_TARGET_OS}.crt1)
+    message(VERBOSE "Skipping ${fq_target_name} as it is not available on ${LIBC_TARGET_OS}.")
+    return()
+  endif()
+
+  cmake_parse_arguments(
+    "GPU_BENCHMARK"
+    "" # No optional arguments
+    "SUITE" # Single value arguments
+    "SRCS;HDRS;DEPENDS;ARGS;ENV;COMPILE_OPTIONS;LINK_LIBRARIES;LOADER_ARGS" # Multi-value arguments
+    ${ARGN}
+  )
+
+  if(NOT GPU_BENCHMARK_SUITE)
+    message(FATAL_ERROR "SUITE not specified for ${fq_target_name}")
+  endif()
+  if(NOT GPU_BENCHMARK_SRCS)
+    message(FATAL_ERROR "The SRCS list for add_gpu_benchmark is missing.")
+  endif()
+
+  get_fq_target_name(${test_name} fq_target_name)
+  get_fq_target_name(${test_name}.libc fq_libc_target_name) # Stores the compiled libc + infrastructure archive to link in
+  get_fq_deps_list(fq_deps_list ${GPU_BENCHMARK_DEPENDS})
+  list(APPEND fq_deps_list
+      # Hermetic tests use the platform's startup object. So, their deps also
+      # have to be collected.
+      libc.startup.${LIBC_TARGET_OS}.crt1
+      # We always add the memory functions objects. This is because the
+      # compiler's codegen can emit calls to the C memory functions.
+      libc.src.string.bcmp
+      libc.src.string.bzero
+      libc.src.string.memcmp
+      libc.src.string.memcpy
+      libc.src.string.memmove
+      libc.src.string.memset
+      libc.src.__support.StringUtil.error_to_string
+  )
+
+  list(REMOVE_DUPLICATES fq_deps_list)
+
+  # TODO: Instead of gathering internal object files from entrypoints,
+  # collect ...
[truncated]

github-actions · 2024-05-13T18:49:43Z

✅ With the latest revision this PR passed the C/C++ code formatter.

jhuber6

Thanks! A few passing comments, I'll look into the benchmark in more detail later.

libc/benchmarks/CMakeLists.txt

libc/benchmarks/gpu/CMakeLists.txt

libc/benchmarks/gpu/LibcGpuBenchmark.cpp

jhuber6 · 2024-05-13T19:30:13Z

libc/benchmarks/gpu/LibcGpuBenchmark.h

+  void Run() override {
+    BenchmarkOptions Options;
+    auto result = benchmark(Options, Func);
+    constexpr auto GREEN = "\033[32m";


Keep in mind that this will need to handle multiple thread sin the future, but I think we can safely restrict this to a single block just to make it easier to manage.

libc/benchmarks/gpu/timing/nvptx/timing.h

libc/benchmarks/gpu/timing/timing.h

libc/benchmarks/gpu/LibcGpuBenchmark.cpp

libc/benchmarks/gpu/LibcGpuBenchmark.h

libc/benchmarks/gpu/timing/nvptx/timing.h

libc/cmake/modules/LLVMLibCTestRules.cmake

libc/benchmarks/gpu/LibcGpuBenchmark.cpp

libc/benchmarks/CMakeLists.txt

jhuber6

Some final nits, looking pretty good overall.

libc/benchmarks/gpu/LibcGpuBenchmark.cpp

jhuber6 · 2024-05-29T00:40:53Z

libc/benchmarks/gpu/LibcGpuBenchmark.cpp

+      break;
+    }
+
+    iterations *= options.scaling_factor;


Nit, could probably move this to the for loop.

which for loop are you referring to? I think my original intent was to increase the number of iterations executed per sample after each sample (i.e. group of iterations) executes. Do you mean moving it to the for loop on line 47?

libc/benchmarks/gpu/LibcGpuBenchmark.cpp

jameshu15869 · 2024-05-30T03:00:27Z

I'm pretty sure the CI fail is caused by my wrong implementation of iterators. Would you be able to review the iterator implementation for anything that seems blatantly wrong? I was a little confused by the implementation of the reverse_iterator and can't seem to get the begin() and end() function calls correct

jhuber6 · 2024-05-30T03:04:21Z

I'm pretty sure the CI fail is caused by my wrong implementation of iterators. Would you be able to review the iterator implementation for anything that seems blatantly wrong? I was a little confused by the implementation of the reverse_iterator and can't seem to get the begin() and end() function calls correct

Can you make the iterators a separate patch? In any case, the iterators should be present in the cpp::Array class, which is pretty much just Ptr and Ptr + Size for the return values of begin, end respectively. Refer to https://en.cppreference.com/w/cpp/language/range-for for what this requires a class to implement to work.

Also might be worth rebasing your branch on main.

nickdesaulniers · 2024-05-30T16:39:48Z

I'm pretty sure the CI fail is caused by my wrong implementation of iterators. Would you be able to review the iterator implementation for anything that seems blatantly wrong? I was a little confused by the implementation of the reverse_iterator and can't seem to get the begin() and end() function calls correct

Make a new PR and send it to me and @gchatelet to review?

- Implements forward iterators for `cpp::fixed_vector` to use in #92009

This reverts commit a5ebf57f198cd79be132854b036f904c3983341d.

jhuber6

I think this is looking pretty good. I'll try some local tests later.

…ild issue

jhuber6

I think this looks good as an initial step. Other libc contributors PTAL if you have any concerns.

@jameshu15869 change the title to reflect its non-draft status.

nickdesaulniers

Minor nit about asm volatile, but LGTM otherwise.

libc/benchmarks/gpu/timing/nvptx/timing.h

libc/benchmarks/gpu/src/math/CMakeLists.txt

PR for adding microbenchmarking infrastructure for NVPTX. `nvlink` cannot perform LTO, so we cannot inline `libc` functions and this function call overhead is not adjusted for during microbenchmarking.

jameshu15869 marked this pull request as draft May 13, 2024 18:46

llvmbot added the libc label May 13, 2024

jhuber6 self-requested a review May 13, 2024 18:48

jameshu15869 force-pushed the nvptx-profiling branch 3 times, most recently from 79772aa to e55e9b6 Compare May 13, 2024 19:23

jhuber6 reviewed May 13, 2024

View reviewed changes

jhuber6 reviewed May 14, 2024

View reviewed changes

jhuber6 reviewed May 22, 2024

View reviewed changes

libc/benchmarks/CMakeLists.txt Show resolved Hide resolved

jameshu15869 force-pushed the nvptx-profiling branch from c87222c to 87a5002 Compare May 23, 2024 16:03

jhuber6 reviewed May 23, 2024

View reviewed changes

libc/benchmarks/CMakeLists.txt Outdated Show resolved Hide resolved

jameshu15869 force-pushed the nvptx-profiling branch from 87a5002 to 087290b Compare May 23, 2024 16:05

jhuber6 reviewed May 29, 2024

View reviewed changes

jameshu15869 force-pushed the nvptx-profiling branch from 8951ce6 to 4e6fdfc Compare May 30, 2024 02:43

jameshu15869 force-pushed the nvptx-profiling branch from 37f1ad5 to 9c47039 Compare May 31, 2024 02:45

jameshu15869 mentioned this pull request May 31, 2024

[libc] Implement forward iterators for libc fixed_vector #93916

Merged

jameshu15869 force-pushed the nvptx-profiling branch from 9c47039 to e777a40 Compare May 31, 2024 03:26

jhuber6 pushed a commit that referenced this pull request Jun 5, 2024

[libc] Implement forward iterators for libc fixed_vector (#93916)

ce8bb9b

- Implements forward iterators for `cpp::fixed_vector` to use in #92009

jameshu15869 added 6 commits June 15, 2024 12:23

initial nvptx microbenchmarking infrastructure

b6b47fb

refactor cmake rules

f8291e9

fix code style

1129ccc

measure walltime, standard deviation, min, and max

5c46009

fixed vector for benchmarks

e50ea99

refactor cmake files

a588fc5

jameshu15869 added 10 commits June 15, 2024 12:23

rename namespace and refactor casts

be303da

repeat overhead measurment outside of loop

a41eb32

switch to using min measurement for overhead

ab6b6ca

fix style

c7c8445

unconditionally add benchmarks in gpu build

c857891

add forward iterator

6073de7

rename logger

9f23d21

Revert "add forward iterator"

46b5e25

This reverts commit a5ebf57f198cd79be132854b036f904c3983341d.

support multithreaded benchmarks

945090f

use for each syntax

4aa5e8b

jameshu15869 force-pushed the nvptx-profiling branch from e777a40 to 4aa5e8b Compare June 15, 2024 18:30

jhuber6 reviewed Jun 15, 2024

View reviewed changes

switch LINK_LIBRARIES argument from optional to multi-value to fix bu…

b93318e

…ild issue

jameshu15869 mentioned this pull request Jun 15, 2024

[libc] Add AMDGPU Timing Utils jameshu15869/llvm-project#1

Closed

jhuber6 marked this pull request as ready for review June 15, 2024 22:12

jhuber6 requested a review from lntue June 15, 2024 22:12

jhuber6 approved these changes Jun 19, 2024

View reviewed changes

jameshu15869 changed the title ~~[libc] NVPTX Profiling Draft~~ [libc] NVPTX Profiling Jun 19, 2024

nickdesaulniers reviewed Jun 20, 2024

View reviewed changes

libc/benchmarks/gpu/timing/nvptx/timing.h Outdated Show resolved Hide resolved

lntue reviewed Jun 20, 2024

View reviewed changes

libc/benchmarks/gpu/src/math/CMakeLists.txt Outdated Show resolved Hide resolved

clean up

cb3b05c

jhuber6 merged commit 02b57de into llvm:main Jun 26, 2024
6 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[libc] NVPTX Profiling #92009

[libc] NVPTX Profiling #92009

jameshu15869 commented May 13, 2024 •

edited

Loading

llvmbot commented May 13, 2024

github-actions bot commented May 13, 2024 •

edited

Loading

jhuber6 left a comment

jhuber6 May 13, 2024

jhuber6 left a comment

jhuber6 May 29, 2024

jameshu15869 May 29, 2024

jameshu15869 commented May 30, 2024

jhuber6 commented May 30, 2024

nickdesaulniers commented May 30, 2024

jhuber6 left a comment

jhuber6 left a comment

nickdesaulniers left a comment

[libc] NVPTX Profiling #92009

[libc] NVPTX Profiling #92009

Conversation

jameshu15869 commented May 13, 2024 • edited Loading

llvmbot commented May 13, 2024

github-actions bot commented May 13, 2024 • edited Loading

jhuber6 left a comment

Choose a reason for hiding this comment

jhuber6 May 13, 2024

Choose a reason for hiding this comment

jhuber6 left a comment

Choose a reason for hiding this comment

jhuber6 May 29, 2024

Choose a reason for hiding this comment

jameshu15869 May 29, 2024

Choose a reason for hiding this comment

jameshu15869 commented May 30, 2024

jhuber6 commented May 30, 2024

nickdesaulniers commented May 30, 2024

jhuber6 left a comment

Choose a reason for hiding this comment

jhuber6 left a comment

Choose a reason for hiding this comment

nickdesaulniers left a comment

Choose a reason for hiding this comment

jameshu15869 commented May 13, 2024 •

edited

Loading

github-actions bot commented May 13, 2024 •

edited

Loading