diff --git a/CMakeLists.txt b/CMakeLists.txt index 81333987bed..2e2cf358e41 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -26,6 +26,7 @@ set(LLVM_MAIN_INCLUDE_DIR ${LLVM_MAIN_SRC_DIR}/include) set(LLVM_BINARY_DIR ${CMAKE_CURRENT_BINARY_DIR}) set(LLVM_TOOLS_BINARY_DIR ${LLVM_BINARY_DIR}/bin) set(LLVM_EXAMPLES_BINARY_DIR ${LLVM_BINARY_DIR}/examples) +set(LLVM_LIBDIR_SUFFIX "" CACHE STRING "Define suffix of library directory name (32/64)" ) set(LLVM_ALL_TARGETS Alpha @@ -186,11 +187,26 @@ if( CMAKE_SIZEOF_VOID_P EQUAL 8 AND NOT WIN32 ) endif( CMAKE_SIZEOF_VOID_P EQUAL 8 AND NOT WIN32 ) if( MSVC ) + # List of valid CRTs for MSVC + set(MSVC_CRT + MD + MDd) + + set(LLVM_USE_CRT "" CACHE STRING "Specify VC++ CRT to use for debug/release configurations.") add_llvm_definitions( -D_CRT_SECURE_NO_DEPRECATE -D_CRT_SECURE_NO_WARNINGS ) add_llvm_definitions( -D_SCL_SECURE_NO_WARNINGS -DCRT_NONSTDC_NO_WARNINGS ) add_llvm_definitions( -D_SCL_SECURE_NO_DEPRECATE ) add_llvm_definitions( -wd4146 -wd4503 -wd4996 -wd4800 -wd4244 -wd4624 ) add_llvm_definitions( -wd4355 -wd4715 -wd4180 -wd4345 -wd4224 ) + + if (NOT ${LLVM_USE_CRT} STREQUAL "") + list(FIND MSVC_CRT ${LLVM_USE_CRT} idx) + if (idx LESS 0) + message(FATAL_ERROR "Invalid value for LLVM_USE_CRT: ${LLVM_USE_CRT}. Valid options are one of: ${MSVC_CRT}") + endif (idx LESS 0) + add_llvm_definitions("/${LLVM_USE_CRT}") + message(STATUS "Using VC++ CRT: ${LLVM_USE_CRT}") + endif (NOT ${LLVM_USE_CRT} STREQUAL "") endif( MSVC ) include_directories( ${LLVM_BINARY_DIR}/include ${LLVM_MAIN_INCLUDE_DIR}) @@ -207,6 +223,8 @@ set(LLVM_COMMON_DEPENDS ${LLVM_COMMON_DEPENDS} ${LLVM_LIBS} ) set(LLVM_TABLEGEN "tblgen" CACHE STRING "Native TableGen executable. Saves building one when cross-compiling.") +# Effective tblgen executable to be used: +set(LLVM_TABLEGEN_EXE ${LLVM_TABLEGEN}) add_subdirectory(utils/TableGen) diff --git a/CREDITS.TXT b/CREDITS.TXT index 0beee20b8d2..76b7296bc2a 100644 --- a/CREDITS.TXT +++ b/CREDITS.TXT @@ -3,7 +3,7 @@ project. If you have contributed a patch or made some other contribution to LLVM, please submit a patch to this file to add yourself, and it will be done! -The list is sorted by name and formatted to allow easy grepping and +The list is sorted by surname and formatted to allow easy grepping and beautification by scripts. The fields are: name (N), email (E), web-address (W), PGP key ID and fingerprint (P), description (D), and snail-mail address (S). @@ -148,10 +148,6 @@ N: Patrick Jenkins E: patjenk@wam.umd.edu D: Nightly Tester -N: Brad Jones -E: kungfoomaster@nondot.org -D: Support for packed types - N: Dale Johannesen E: dalej@apple.com D: ARM constant islands improvements @@ -160,6 +156,10 @@ D: Rewrite X87 back end D: Use APFloat for floating point constants widely throughout compiler D: Implement X87 long double +N: Brad Jones +E: kungfoomaster@nondot.org +D: Support for packed types + N: Eric Kidd W: http://randomhacks.net/ D: llvm-config script @@ -231,6 +231,13 @@ N: Scott Michel E: scottm@aero.org D: Added STI Cell SPU backend. +N: Edward O'Callaghan +E: eocallaghan@auroraux.org +W: http://www.auroraux.org +D: Add Clang support with various other improvements to utils/NewNightlyTest.pl +D: Fix and maintain Solaris & AuroraUX support for llvm, various build warnings +D: and error clean ups. + N: Morten Ofstad E: morten@hue.no D: Visual C++ compatibility fixes @@ -266,6 +273,10 @@ N: Arnold Schwaighofer E: arnold.schwaighofer@gmail.com D: Tail call optimization for the x86 backend +N: Shantonu Sen +E: ssen@apple.com +D: Miscellaneous bug fixes + N: Anand Shukla E: ashukla@cs.uiuc.edu D: The `paths' pass @@ -290,8 +301,4 @@ D: Thread Local Storage implementation N: Bill Wendling E: isanbard@gmail.com -D: Machine LICM -D: Darwin exception handling -D: MMX & SSSE3 instructions -D: SPEC2006 support - +D: Bunches of stuff diff --git a/cmake/modules/AddLLVM.cmake b/cmake/modules/AddLLVM.cmake index a21ed20d3ba..e5256746bcc 100755 --- a/cmake/modules/AddLLVM.cmake +++ b/cmake/modules/AddLLVM.cmake @@ -10,8 +10,8 @@ macro(add_llvm_library name) add_dependencies( ${name} ${LLVM_COMMON_DEPENDS} ) endif( LLVM_COMMON_DEPENDS ) install(TARGETS ${name} - LIBRARY DESTINATION lib - ARCHIVE DESTINATION lib) + LIBRARY DESTINATION lib${LLVM_LIBDIR_SUFFIX} + ARCHIVE DESTINATION lib${LLVM_LIBDIR_SUFFIX}) endmacro(add_llvm_library name) diff --git a/cmake/modules/AddPartiallyLinkedObject.cmake b/cmake/modules/AddPartiallyLinkedObject.cmake index d20666dd79f..0f92455eacd 100755 --- a/cmake/modules/AddPartiallyLinkedObject.cmake +++ b/cmake/modules/AddPartiallyLinkedObject.cmake @@ -38,5 +38,5 @@ macro(add_partially_linked_object lib) set( llvm_lib_targets ${llvm_lib_targets} ${tnplo} PARENT_SCOPE ) endif( ) install(FILES ${pll} - DESTINATION lib) + DESTINATION lib${LLVM_LIBDIR_SUFFIX}) endmacro(add_partially_linked_object lib) diff --git a/cmake/modules/CrossCompileLLVM.cmake b/cmake/modules/CrossCompileLLVM.cmake index f638d67a30d..138ff0e9fe6 100644 --- a/cmake/modules/CrossCompileLLVM.cmake +++ b/cmake/modules/CrossCompileLLVM.cmake @@ -1,7 +1,7 @@ if( ${LLVM_TABLEGEN} STREQUAL "tblgen" ) set(CX_NATIVE_TG_DIR "${CMAKE_BINARY_DIR}/native") - set(LLVM_TABLEGEN "${CX_NATIVE_TG_DIR}/bin/tblgen") + set(LLVM_TABLEGEN_EXE "${CX_NATIVE_TG_DIR}/bin/tblgen") add_custom_command(OUTPUT ${CX_NATIVE_TG_DIR} COMMAND ${CMAKE_COMMAND} -E make_directory ${CX_NATIVE_TG_DIR} @@ -13,12 +13,12 @@ if( ${LLVM_TABLEGEN} STREQUAL "tblgen" ) DEPENDS ${CX_NATIVE_TG_DIR} COMMENT "Configuring native TableGen...") - add_custom_command(OUTPUT ${LLVM_TABLEGEN} + add_custom_command(OUTPUT ${LLVM_TABLEGEN_EXE} COMMAND ${CMAKE_BUILD_TOOL} DEPENDS ${CX_NATIVE_TG_DIR}/CMakeCache.txt WORKING_DIRECTORY ${CX_NATIVE_TG_DIR}/utils/TableGen COMMENT "Building native TableGen...") - add_custom_target(NativeTableGen DEPENDS ${LLVM_TABLEGEN}) + add_custom_target(NativeTableGen DEPENDS ${LLVM_TABLEGEN_EXE}) add_dependencies(tblgen NativeTableGen) diff --git a/cmake/modules/TableGen.cmake b/cmake/modules/TableGen.cmake index adb22c726b9..16c732b0893 100644 --- a/cmake/modules/TableGen.cmake +++ b/cmake/modules/TableGen.cmake @@ -6,11 +6,11 @@ macro(tablegen ofn) file(GLOB all_tds "*.td") add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/${ofn}.tmp - COMMAND ${LLVM_TABLEGEN} ${ARGN} -I ${CMAKE_CURRENT_SOURCE_DIR} + COMMAND ${LLVM_TABLEGEN_EXE} ${ARGN} -I ${CMAKE_CURRENT_SOURCE_DIR} -I ${LLVM_MAIN_SRC_DIR}/lib/Target -I ${LLVM_MAIN_INCLUDE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/${LLVM_TARGET_DEFINITIONS} -o ${CMAKE_CURRENT_BINARY_DIR}/${ofn}.tmp - DEPENDS ${LLVM_TABLEGEN} ${all_tds} + DEPENDS tblgen ${all_tds} COMMENT "Building ${ofn}.tmp..." ) add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/${ofn} diff --git a/docs/CMake.html b/docs/CMake.html index ac3b57a8489..1f50d397f00 100644 --- a/docs/CMake.html +++ b/docs/CMake.html @@ -217,6 +217,11 @@
Path where LLVM will be installed if "make install" is invoked or the "INSTALL" target is built.
+
LLVM_LIBDIR_SUFFIX:STRING
+
Extra suffix to append to the directory where libraries are to + be installed. On a 64-bit architecture, one could use + -DLLVM_LIBDIR_SUFFIX=64 to install libraries to /usr/lib64.
+
CMAKE_C_FLAGS:STRING
Extra flags to use when compiling C source files.
@@ -296,7 +301,13 @@
-

TODO

+

See this + wiki page for generic instructions on how to cross-compile + with CMake. It goes into detailed explanations and may seem + daunting, but it is not. On the wiki page there are several + examples including toolchain files. Go directly to + this + section for a quick solution.

diff --git a/docs/LangRef.html b/docs/LangRef.html index 89d4f936569..897654df956 100644 --- a/docs/LangRef.html +++ b/docs/LangRef.html @@ -1091,19 +1091,27 @@ stack before the local variables that's checked upon return from the function to see if it has been overwritten. A heuristic is used to determine if a function needs stack protectors or not. -

If a function that has an ssp attribute is inlined into a function +

If a function that has an ssp attribute is inlined into a function that doesn't have an ssp attribute, then the resulting function will -have an ssp attribute.

+have an ssp attribute.
sspreq
This attribute indicates that the function should always emit a stack smashing protector. This overrides the ssp function attribute. -

If a function that has an sspreq attribute is inlined into a +If a function that has an sspreq attribute is inlined into a function that doesn't have an sspreq attribute or which has an ssp attribute, then the resulting function will have -an sspreq attribute.

+an sspreq attribute. + +
noredzone
+
This attribute indicates that the code generator should not enforce red zone +mandated by target specific ABI.
+ +
noimplicitfloat
+
This attributes disables implicit floating point instructions.
+ @@ -1177,6 +1185,9 @@ aspect of the data layout. The specifications accepted are as follows:

asize:abi:pref
This specifies the alignment for an aggregate type of a given bit size.
+
ssize:abi:pref
+
This specifies the alignment for a stack object of a given bit + size.

When constructing the data layout for a given target, LLVM starts with a default set of specifications which are then (possibly) overriden by the @@ -1196,6 +1207,7 @@ are given in this list:

  • v64:64:64 - 64-bit vector is 64-bit aligned
  • v128:128:128 - 128-bit vector is 128-bit aligned
  • a0:0:1 - aggregates are 8-bit aligned
    • +
  • s0:64:64 - stack objects are 64-bit aligned

    • When LLVM is determining the alignment for a given type, it uses the following rules:

    @@ -7209,7 +7221,7 @@ declare void @llvm.stackprotector( i8* <guard>, i8** <slot> ) Chris Lattner
    The LLVM Compiler Infrastructure
    - Last modified: $Date: 2009-06-05 00:49:04 +0200 (Fri, 05 Jun 2009) $ + Last modified: $Date: 2009-06-12 21:45:19 +0200 (Fri, 12 Jun 2009) $ diff --git a/docs/TableGenFundamentals.html b/docs/TableGenFundamentals.html index f477b474fb6..48c60b90274 100644 --- a/docs/TableGenFundamentals.html +++ b/docs/TableGenFundamentals.html @@ -371,8 +371,11 @@ supported include:

    string value
    [{ ... }]
    code fragment
    -
    [ X, Y, Z ]
    -
    list value.
    +
    [ X, Y, Z ]
    +
    list value. is the type of the list +element and is usually optional. In rare cases, +TableGen is unable to deduce the element type in +which case the user must specify it explicitly.
    { a, b, c }
    initializer for a "bits<3>" value
    value
    @@ -778,7 +781,7 @@ This should highlight the APIs in TableGen/Record.h.

    Chris Lattner
    LLVM Compiler Infrastructure
    - Last modified: $Date: 2009-05-19 00:14:45 +0200 (Tue, 19 May 2009) $ + Last modified: $Date: 2009-06-09 20:31:17 +0200 (Tue, 09 Jun 2009) $ diff --git a/include/llvm/Analysis/ScalarEvolution.h b/include/llvm/Analysis/ScalarEvolution.h index e3d429934cc..41725be1ca3 100644 --- a/include/llvm/Analysis/ScalarEvolution.h +++ b/include/llvm/Analysis/ScalarEvolution.h @@ -393,6 +393,7 @@ namespace llvm { SCEVHandle getTruncateExpr(const SCEVHandle &Op, const Type *Ty); SCEVHandle getZeroExtendExpr(const SCEVHandle &Op, const Type *Ty); SCEVHandle getSignExtendExpr(const SCEVHandle &Op, const Type *Ty); + SCEVHandle getAnyExtendExpr(const SCEVHandle &Op, const Type *Ty); SCEVHandle getAddExpr(std::vector &Ops); SCEVHandle getAddExpr(const SCEVHandle &LHS, const SCEVHandle &RHS) { std::vector Ops; @@ -465,6 +466,12 @@ namespace llvm { /// it is sign extended. The conversion must not be narrowing. SCEVHandle getNoopOrSignExtend(const SCEVHandle &V, const Type *Ty); + /// getNoopOrAnyExtend - Return a SCEV corresponding to a conversion of + /// the input value to the specified type. If the type must be extended, + /// it is extended with unspecified bits. The conversion must not be + /// narrowing. + SCEVHandle getNoopOrAnyExtend(const SCEVHandle &V, const Type *Ty); + /// getTruncateOrNoop - Return a SCEV corresponding to a conversion of the /// input value to the specified type. The conversion must not be /// widening. diff --git a/include/llvm/Attributes.h b/include/llvm/Attributes.h index a594e3223d4..134e3502028 100644 --- a/include/llvm/Attributes.h +++ b/include/llvm/Attributes.h @@ -61,7 +61,8 @@ const Attributes NoImplicitFloat = 1<<23; /// disable implicit floating point /// @brief Attributes that only apply to function parameters. const Attributes ParameterOnly = ByVal | Nest | StructRet | NoCapture; -/// @brief Attributes that only apply to function. +/// @brief Attributes that may be applied to the function itself. These cannot +/// be used on return values or function parameters. const Attributes FunctionOnly = NoReturn | NoUnwind | ReadNone | ReadOnly | NoInline | AlwaysInline | OptimizeForSize | StackProtect | StackProtectReq | NoRedZone | NoImplicitFloat; @@ -186,7 +187,7 @@ public: /// getFnAttributes - The function attributes are returned. Attributes getFnAttributes() const { - return getAttributes(~0); + return getAttributes(~0U); } /// paramHasAttr - Return true if the specified parameter index has the diff --git a/include/llvm/CodeGen/BinaryObject.h b/include/llvm/CodeGen/BinaryObject.h new file mode 100644 index 00000000000..0780cd6ab4f --- /dev/null +++ b/include/llvm/CodeGen/BinaryObject.h @@ -0,0 +1,325 @@ +//===-- llvm/CodeGen/BinaryObject.h - Binary Object. -----------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines a Binary Object Aka. "blob" for holding data from code +// generators, ready for data to the object module code writters. +// +//===----------------------------------------------------------------------===// + +#ifndef LLVM_CODEGEN_BINARYOBJECT_H +#define LLVM_CODEGEN_BINARYOBJECT_H + +#include +#include + +namespace llvm { + +class MachineRelocation; +typedef std::vector BinaryData; + +class BinaryObject { +protected: + std::string Name; + bool IsLittleEndian; + bool Is64Bit; + BinaryData Data; + std::vector Relocations; + +public: + /// Constructors and destructor + BinaryObject() {} + + BinaryObject(bool isLittleEndian, bool is64Bit) + : IsLittleEndian(isLittleEndian), Is64Bit(is64Bit) {} + + BinaryObject(const std::string &name, bool isLittleEndian, bool is64Bit) + : Name(name), IsLittleEndian(isLittleEndian), Is64Bit(is64Bit) {} + + ~BinaryObject() {} + + /// getName - get name of BinaryObject + inline std::string getName() const { return Name; } + + /// get size of binary data + size_t size() const { + return Data.size(); + } + + /// get binary data + BinaryData& getData() { + return Data; + } + + /// get machine relocations + const std::vector& getRelocations() const { + return Relocations; + } + + /// emitByte - This callback is invoked when a byte needs to be + /// written to the data stream. + inline void emitByte(uint8_t B) { + Data.push_back(B); + } + + /// emitWord16 - This callback is invoked when a 16-bit word needs to be + /// written to the data stream in correct endian format and correct size. + inline void emitWord16(uint16_t W) { + if (IsLittleEndian) + emitWord16LE(W); + else + emitWord16BE(W); + } + + /// emitWord16LE - This callback is invoked when a 16-bit word needs to be + /// written to the data stream in correct endian format and correct size. + inline void emitWord16LE(uint16_t W) { + Data.push_back((W >> 0) & 255); + Data.push_back((W >> 8) & 255); + } + + /// emitWord16BE - This callback is invoked when a 16-bit word needs to be + /// written to the data stream in correct endian format and correct size. + inline void emitWord16BE(uint16_t W) { + Data.push_back((W >> 8) & 255); + Data.push_back((W >> 0) & 255); + } + + /// emitWord - This callback is invoked when a word needs to be + /// written to the data stream in correct endian format and correct size. + inline void emitWord(uint64_t W) { + if (!Is64Bit) + emitWord32(W); + else + emitWord64(W); + } + + /// emitWord32 - This callback is invoked when a 32-bit word needs to be + /// written to the data stream in correct endian format. + inline void emitWord32(uint32_t W) { + if (IsLittleEndian) + emitWordLE(W); + else + emitWordBE(W); + } + + /// emitWord64 - This callback is invoked when a 32-bit word needs to be + /// written to the data stream in correct endian format. + inline void emitWord64(uint64_t W) { + if (IsLittleEndian) + emitDWordLE(W); + else + emitDWordBE(W); + } + + /// emitWordLE - This callback is invoked when a 32-bit word needs to be + /// written to the data stream in little-endian format. + inline void emitWordLE(uint32_t W) { + Data.push_back((W >> 0) & 255); + Data.push_back((W >> 8) & 255); + Data.push_back((W >> 16) & 255); + Data.push_back((W >> 24) & 255); + } + + /// emitWordBE - This callback is invoked when a 32-bit word needs to be + /// written to the data stream in big-endian format. + /// + inline void emitWordBE(uint32_t W) { + Data.push_back((W >> 24) & 255); + Data.push_back((W >> 16) & 255); + Data.push_back((W >> 8) & 255); + Data.push_back((W >> 0) & 255); + } + + /// emitDWordLE - This callback is invoked when a 64-bit word needs to be + /// written to the data stream in little-endian format. + inline void emitDWordLE(uint64_t W) { + Data.push_back(unsigned(W >> 0) & 255); + Data.push_back(unsigned(W >> 8) & 255); + Data.push_back(unsigned(W >> 16) & 255); + Data.push_back(unsigned(W >> 24) & 255); + Data.push_back(unsigned(W >> 32) & 255); + Data.push_back(unsigned(W >> 40) & 255); + Data.push_back(unsigned(W >> 48) & 255); + Data.push_back(unsigned(W >> 56) & 255); + } + + /// emitDWordBE - This callback is invoked when a 64-bit word needs to be + /// written to the data stream in big-endian format. + inline void emitDWordBE(uint64_t W) { + Data.push_back(unsigned(W >> 56) & 255); + Data.push_back(unsigned(W >> 48) & 255); + Data.push_back(unsigned(W >> 40) & 255); + Data.push_back(unsigned(W >> 32) & 255); + Data.push_back(unsigned(W >> 24) & 255); + Data.push_back(unsigned(W >> 16) & 255); + Data.push_back(unsigned(W >> 8) & 255); + Data.push_back(unsigned(W >> 0) & 255); + } + + /// fixByte - This callback is invoked when a byte needs to be + /// fixup the buffer. + inline void fixByte(uint8_t B, uint32_t offset) { + Data[offset] = B; + } + + /// fixWord16 - This callback is invoked when a 16-bit word needs to + /// fixup the data stream in correct endian format. + inline void fixWord16(uint16_t W, uint32_t offset) { + if (IsLittleEndian) + fixWord16LE(W, offset); + else + fixWord16BE(W, offset); + } + + /// emitWord16LE - This callback is invoked when a 16-bit word needs to + /// fixup the data stream in little endian format. + inline void fixWord16LE(uint16_t W, uint32_t offset) { + Data[offset++] = W & 255; + Data[offset] = (W >> 8) & 255; + } + + /// fixWord16BE - This callback is invoked when a 16-bit word needs to + /// fixup data stream in big endian format. + inline void fixWord16BE(uint16_t W, uint32_t offset) { + Data[offset++] = (W >> 8) & 255; + Data[offset] = W & 255; + } + + /// emitWord - This callback is invoked when a word needs to + /// fixup the data in correct endian format and correct size. + inline void fixWord(uint64_t W, uint32_t offset) { + if (!Is64Bit) + fixWord32(W, offset); + else + fixWord64(W, offset); + } + + /// fixWord32 - This callback is invoked when a 32-bit word needs to + /// fixup the data in correct endian format. + inline void fixWord32(uint32_t W, uint32_t offset) { + if (IsLittleEndian) + fixWord32LE(W, offset); + else + fixWord32BE(W, offset); + } + + /// fixWord32LE - This callback is invoked when a 32-bit word needs to + /// fixup the data in little endian format. + inline void fixWord32LE(uint32_t W, uint32_t offset) { + Data[offset++] = W & 255; + Data[offset++] = (W >> 8) & 255; + Data[offset++] = (W >> 16) & 255; + Data[offset] = (W >> 24) & 255; + } + + /// fixWord32BE - This callback is invoked when a 32-bit word needs to + /// fixup the data in big endian format. + inline void fixWord32BE(uint32_t W, uint32_t offset) { + Data[offset++] = (W >> 24) & 255; + Data[offset++] = (W >> 16) & 255; + Data[offset++] = (W >> 8) & 255; + Data[offset] = W & 255; + } + + /// fixWord64 - This callback is invoked when a 64-bit word needs to + /// fixup the data in correct endian format. + inline void fixWord64(uint64_t W, uint32_t offset) { + if (IsLittleEndian) + fixWord64LE(W, offset); + else + fixWord64BE(W, offset); + } + + /// fixWord64BE - This callback is invoked when a 64-bit word needs to + /// fixup the data in little endian format. + inline void fixWord64LE(uint64_t W, uint32_t offset) { + Data[offset++] = W & 255; + Data[offset++] = (W >> 8) & 255; + Data[offset++] = (W >> 16) & 255; + Data[offset++] = (W >> 24) & 255; + Data[offset++] = (W >> 32) & 255; + Data[offset++] = (W >> 40) & 255; + Data[offset++] = (W >> 48) & 255; + Data[offset] = (W >> 56) & 255; + } + + /// fixWord64BE - This callback is invoked when a 64-bit word needs to + /// fixup the data in big endian format. + inline void fixWord64BE(uint64_t W, uint32_t offset) { + Data[offset++] = (W >> 56) & 255; + Data[offset++] = (W >> 48) & 255; + Data[offset++] = (W >> 40) & 255; + Data[offset++] = (W >> 32) & 255; + Data[offset++] = (W >> 24) & 255; + Data[offset++] = (W >> 16) & 255; + Data[offset++] = (W >> 8) & 255; + Data[offset] = W & 255; + } + + /// emitAlignment - Pad the data to the specified alignment. + void emitAlignment(unsigned Alignment) { + if (Alignment <= 1) return; + unsigned PadSize = -Data.size() & (Alignment-1); + for (unsigned i = 0; i>= 7; + if (Value) Byte |= 0x80; + emitByte(Byte); + } while (Value); + } + + /// emitSLEB128Bytes - This callback is invoked when a SLEB128 needs to be + /// written to the data stream. + void emitSLEB128Bytes(int64_t Value) { + int Sign = Value >> (8 * sizeof(Value) - 1); + bool IsMore; + + do { + unsigned char Byte = Value & 0x7f; + Value >>= 7; + IsMore = Value != Sign || ((Byte ^ Sign) & 0x40) != 0; + if (IsMore) Byte |= 0x80; + emitByte(Byte); + } while (IsMore); + } + + /// emitString - This callback is invoked when a String needs to be + /// written to the data stream. + void emitString(const std::string &String) { + for (unsigned i = 0, N = static_cast(String.size()); i> 0); *CurBufferPtr++ = (uint8_t)(W >> 8); @@ -111,7 +111,7 @@ public: /// emitWordBE - This callback is invoked when a 32-bit word needs to be /// written to the output stream in big-endian format. /// - void emitWordBE(unsigned W) { + void emitWordBE(uint32_t W) { if (4 <= BufferEnd-CurBufferPtr) { *CurBufferPtr++ = (uint8_t)(W >> 24); *CurBufferPtr++ = (uint8_t)(W >> 16); @@ -176,7 +176,7 @@ public: /// emitULEB128Bytes - This callback is invoked when a ULEB128 needs to be /// written to the output stream. - void emitULEB128Bytes(unsigned Value) { + void emitULEB128Bytes(uint64_t Value) { do { uint8_t Byte = Value & 0x7f; Value >>= 7; @@ -187,7 +187,7 @@ public: /// emitSLEB128Bytes - This callback is invoked when a SLEB128 needs to be /// written to the output stream. - void emitSLEB128Bytes(int32_t Value) { + void emitSLEB128Bytes(int64_t Value) { int32_t Sign = Value >> (8 * sizeof(Value) - 1); bool IsMore; @@ -212,7 +212,7 @@ public: } /// emitInt32 - Emit a int32 directive. - void emitInt32(int32_t Value) { + void emitInt32(uint32_t Value) { if (4 <= BufferEnd-CurBufferPtr) { *((uint32_t*)CurBufferPtr) = Value; CurBufferPtr += 4; diff --git a/include/llvm/CodeGen/LazyLiveness.h b/include/llvm/CodeGen/LazyLiveness.h new file mode 100644 index 00000000000..82e4a153d51 --- /dev/null +++ b/include/llvm/CodeGen/LazyLiveness.h @@ -0,0 +1,63 @@ +//===- LazyLiveness.h - Lazy, CFG-invariant liveness information ----------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This pass implements a lazy liveness analysis as per "Fast Liveness Checking +// for SSA-form Programs," by Boissinot, et al. +// +//===----------------------------------------------------------------------===// + +#ifndef LLVM_CODEGEN_LAZYLIVENESS_H +#define LLVM_CODEGEN_LAZYLIVENESS_H + +#include "llvm/CodeGen/MachineFunctionPass.h" +#include "llvm/CodeGen/MachineDominators.h" +#include "llvm/ADT/DenseMap.h" +#include "llvm/ADT/DenseSet.h" +#include "llvm/ADT/SparseBitVector.h" +#include + +namespace llvm { + +class MachineRegisterInfo; + +class LazyLiveness : public MachineFunctionPass { +public: + static char ID; // Pass identification, replacement for typeid + LazyLiveness() : MachineFunctionPass(&ID) { } + + void getAnalysisUsage(AnalysisUsage &AU) const { + AU.setPreservesAll(); + AU.addRequired(); + } + + bool runOnMachineFunction(MachineFunction &mf); + + bool vregLiveIntoMBB(unsigned vreg, MachineBasicBlock* MBB); + +private: + void computeBackedgeChain(MachineFunction& mf, MachineBasicBlock* MBB); + + typedef std::pair edge_t; + + MachineRegisterInfo* MRI; + + DenseMap preorder; + std::vector rev_preorder; + DenseMap > rv; + DenseMap > tv; + DenseSet backedges; + SparseBitVector<128> backedge_source; + SparseBitVector<128> backedge_target; + SparseBitVector<128> calculated; +}; + +} + +#endif + diff --git a/include/llvm/CodeGen/MachineCodeEmitter.h b/include/llvm/CodeGen/MachineCodeEmitter.h index aaa41a48cb2..eb1ea2dc56b 100644 --- a/include/llvm/CodeGen/MachineCodeEmitter.h +++ b/include/llvm/CodeGen/MachineCodeEmitter.h @@ -104,7 +104,7 @@ public: /// emitWordLE - This callback is invoked when a 32-bit word needs to be /// written to the output stream in little-endian format. /// - void emitWordLE(unsigned W) { + void emitWordLE(uint32_t W) { if (4 <= BufferEnd-CurBufferPtr) { *CurBufferPtr++ = (uint8_t)(W >> 0); *CurBufferPtr++ = (uint8_t)(W >> 8); @@ -118,7 +118,7 @@ public: /// emitWordBE - This callback is invoked when a 32-bit word needs to be /// written to the output stream in big-endian format. /// - void emitWordBE(unsigned W) { + void emitWordBE(uint32_t W) { if (4 <= BufferEnd-CurBufferPtr) { *CurBufferPtr++ = (uint8_t)(W >> 24); *CurBufferPtr++ = (uint8_t)(W >> 16); @@ -183,7 +183,7 @@ public: /// emitULEB128Bytes - This callback is invoked when a ULEB128 needs to be /// written to the output stream. - void emitULEB128Bytes(unsigned Value) { + void emitULEB128Bytes(uint64_t Value) { do { uint8_t Byte = Value & 0x7f; Value >>= 7; @@ -194,8 +194,8 @@ public: /// emitSLEB128Bytes - This callback is invoked when a SLEB128 needs to be /// written to the output stream. - void emitSLEB128Bytes(int32_t Value) { - int32_t Sign = Value >> (8 * sizeof(Value) - 1); + void emitSLEB128Bytes(uint64_t Value) { + uint64_t Sign = Value >> (8 * sizeof(Value) - 1); bool IsMore; do { diff --git a/include/llvm/ExecutionEngine/ExecutionEngine.h b/include/llvm/ExecutionEngine/ExecutionEngine.h index 0d171f6764d..170e18477a7 100644 --- a/include/llvm/ExecutionEngine/ExecutionEngine.h +++ b/include/llvm/ExecutionEngine/ExecutionEngine.h @@ -243,7 +243,7 @@ public: } // The JIT overrides a version that actually does this. - virtual void runJITOnFunction(Function *F, MachineCodeInfo *MCI = 0) { } + virtual void runJITOnFunction(Function *, MachineCodeInfo * = 0) { } /// getGlobalValueAtAddress - Return the LLVM global value object that starts /// at the specified address. diff --git a/include/llvm/Function.h b/include/llvm/Function.h index ccc006cfcfb..228ef9440e0 100644 --- a/include/llvm/Function.h +++ b/include/llvm/Function.h @@ -395,6 +395,10 @@ public: /// including any contained basic blocks. /// void dropAllReferences(); + + /// hasAddressTaken - returns true if there are any uses of this function + /// other than direct calls or invokes to it. + bool hasAddressTaken() const; }; inline ValueSymbolTable * diff --git a/include/llvm/LinkAllPasses.h b/include/llvm/LinkAllPasses.h index edd27223aa8..0fb837da772 100644 --- a/include/llvm/LinkAllPasses.h +++ b/include/llvm/LinkAllPasses.h @@ -127,6 +127,7 @@ namespace { (void) llvm::createPrintModulePass(0); (void) llvm::createPrintFunctionPass("", 0); (void) llvm::createDbgInfoPrinterPass(); + (void) llvm::createPartialInliningPass(); (void)new llvm::IntervalPartition(); (void)new llvm::FindUsedTypes(); diff --git a/include/llvm/Target/TargetELFWriterInfo.h b/include/llvm/Target/TargetELFWriterInfo.h index e266a71f8c8..f7e3392577d 100644 --- a/include/llvm/Target/TargetELFWriterInfo.h +++ b/include/llvm/Target/TargetELFWriterInfo.h @@ -15,15 +15,21 @@ #define LLVM_TARGET_TARGETELFWRITERINFO_H namespace llvm { + class Function; + class TargetData; + class TargetMachine; //===--------------------------------------------------------------------===// // TargetELFWriterInfo //===--------------------------------------------------------------------===// class TargetELFWriterInfo { + protected: // EMachine - This field is the target specific value to emit as the // e_machine member of the ELF header. unsigned short EMachine; + TargetMachine &TM; + bool is64Bit, isLittleEndian; public: // Machine architectures @@ -44,10 +50,39 @@ namespace llvm { EM_X86_64 = 62 // AMD64 }; - explicit TargetELFWriterInfo(MachineType machine) : EMachine(machine) {} - virtual ~TargetELFWriterInfo() {} + // ELF File classes + enum { + ELFCLASS32 = 1, // 32-bit object file + ELFCLASS64 = 2 // 64-bit object file + }; + + // ELF Endianess + enum { + ELFDATA2LSB = 1, // Little-endian object file + ELFDATA2MSB = 2 // Big-endian object file + }; + + explicit TargetELFWriterInfo(TargetMachine &tm); + virtual ~TargetELFWriterInfo(); unsigned short getEMachine() const { return EMachine; } + unsigned getEFlags() const { return 0; } + unsigned getEIClass() const { return is64Bit ? ELFCLASS64 : ELFCLASS32; } + unsigned getEIData() const { + return isLittleEndian ? ELFDATA2LSB : ELFDATA2MSB; + } + + /// ELF Header and ELF Section Header Info + unsigned getHdrSize() const { return is64Bit ? 64 : 52; } + unsigned getSHdrSize() const { return is64Bit ? 64 : 40; } + + /// Symbol Table Info + unsigned getSymTabEntrySize() const { return is64Bit ? 24 : 16; } + unsigned getSymTabAlignment() const { return is64Bit ? 8 : 4; } + + /// getFunctionAlignment - Returns the alignment for function 'F', targets + /// with different alignment constraints should overload this method + virtual unsigned getFunctionAlignment(const Function *F) const; }; } // end llvm namespace diff --git a/include/llvm/Target/TargetLowering.h b/include/llvm/Target/TargetLowering.h index 327af275114..47dcc6c8e48 100644 --- a/include/llvm/Target/TargetLowering.h +++ b/include/llvm/Target/TargetLowering.h @@ -350,7 +350,7 @@ public: LegalizeAction getOperationAction(unsigned Op, MVT VT) const { if (VT.isExtended()) return Expand; assert(Op < array_lengthof(OpActions) && - (unsigned)VT.getSimpleVT() < sizeof(OpActions[0])*4 && + (unsigned)VT.getSimpleVT() < sizeof(OpActions[0])*8 && "Table isn't big enough!"); return (LegalizeAction)((OpActions[Op] >> (2*VT.getSimpleVT())) & 3); } @@ -417,11 +417,10 @@ public: /// for it. LegalizeAction getIndexedLoadAction(unsigned IdxMode, MVT VT) const { - assert(IdxMode < array_lengthof(IndexedModeActions[0]) && - (unsigned)VT.getSimpleVT() < sizeof(IndexedModeActions[0][0])*4 && + assert( IdxMode < array_lengthof(IndexedModeActions[0][0]) && + ((unsigned)VT.getSimpleVT()) < MVT::LAST_VALUETYPE && "Table isn't big enough!"); - return (LegalizeAction)((IndexedModeActions[0][IdxMode] >> - (2*VT.getSimpleVT())) & 3); + return (LegalizeAction)((IndexedModeActions[(unsigned)VT.getSimpleVT()][0][IdxMode])); } /// isIndexedLoadLegal - Return true if the specified indexed load is legal @@ -438,11 +437,10 @@ public: /// for it. LegalizeAction getIndexedStoreAction(unsigned IdxMode, MVT VT) const { - assert(IdxMode < array_lengthof(IndexedModeActions[1]) && - (unsigned)VT.getSimpleVT() < sizeof(IndexedModeActions[1][0])*4 && + assert(IdxMode < array_lengthof(IndexedModeActions[0][1]) && + (unsigned)VT.getSimpleVT() < MVT::LAST_VALUETYPE && "Table isn't big enough!"); - return (LegalizeAction)((IndexedModeActions[1][IdxMode] >> - (2*VT.getSimpleVT())) & 3); + return (LegalizeAction)((IndexedModeActions[(unsigned)VT.getSimpleVT()][1][IdxMode])); } /// isIndexedStoreLegal - Return true if the specified indexed load is legal @@ -942,7 +940,7 @@ protected: /// with the specified type and indicate what to do about it. void setOperationAction(unsigned Op, MVT VT, LegalizeAction Action) { - assert((unsigned)VT.getSimpleVT() < sizeof(OpActions[0])*4 && + assert((unsigned)VT.getSimpleVT() < sizeof(OpActions[0])*8 && Op < array_lengthof(OpActions) && "Table isn't big enough!"); OpActions[Op] &= ~(uint64_t(3UL) << VT.getSimpleVT()*2); OpActions[Op] |= (uint64_t)Action << VT.getSimpleVT()*2; @@ -978,11 +976,10 @@ protected: /// TargetLowering.cpp void setIndexedLoadAction(unsigned IdxMode, MVT VT, LegalizeAction Action) { - assert((unsigned)VT.getSimpleVT() < sizeof(IndexedModeActions[0])*4 && - IdxMode < array_lengthof(IndexedModeActions[0]) && + assert((unsigned)VT.getSimpleVT() < MVT::LAST_VALUETYPE && + IdxMode < array_lengthof(IndexedModeActions[0][0]) && "Table isn't big enough!"); - IndexedModeActions[0][IdxMode] &= ~(uint64_t(3UL) << VT.getSimpleVT()*2); - IndexedModeActions[0][IdxMode] |= (uint64_t)Action << VT.getSimpleVT()*2; + IndexedModeActions[(unsigned)VT.getSimpleVT()][0][IdxMode] = (uint8_t)Action; } /// setIndexedStoreAction - Indicate that the specified indexed store does or @@ -991,11 +988,10 @@ protected: /// TargetLowering.cpp void setIndexedStoreAction(unsigned IdxMode, MVT VT, LegalizeAction Action) { - assert((unsigned)VT.getSimpleVT() < sizeof(IndexedModeActions[1][0])*4 && - IdxMode < array_lengthof(IndexedModeActions[1]) && + assert((unsigned)VT.getSimpleVT() < MVT::LAST_VALUETYPE && + IdxMode < array_lengthof(IndexedModeActions[0][1] ) && "Table isn't big enough!"); - IndexedModeActions[1][IdxMode] &= ~(uint64_t(3UL) << VT.getSimpleVT()*2); - IndexedModeActions[1][IdxMode] |= (uint64_t)Action << VT.getSimpleVT()*2; + IndexedModeActions[(unsigned)VT.getSimpleVT()][1][IdxMode] = (uint8_t)Action; } /// setConvertAction - Indicate that the specified conversion does or does @@ -1581,10 +1577,13 @@ private: /// indicates how instruction selection should deal with the store. uint64_t TruncStoreActions[MVT::LAST_VALUETYPE]; - /// IndexedModeActions - For each indexed mode and each value type, keep a - /// pair of LegalizeAction that indicates how instruction selection should - /// deal with the load / store. - uint64_t IndexedModeActions[2][ISD::LAST_INDEXED_MODE]; + /// IndexedModeActions - For each indexed mode and each value type, + /// keep a pair of LegalizeAction that indicates how instruction + /// selection should deal with the load / store. The first + /// dimension is now the value_type for the reference. The second + /// dimension is the load [0] vs. store[1]. The third dimension + /// represents the various modes for load store. + uint8_t IndexedModeActions[MVT::LAST_VALUETYPE][2][ISD::LAST_INDEXED_MODE]; /// ConvertActions - For each conversion from source type to destination type, /// keep a LegalizeAction that indicates how instruction selection should diff --git a/include/llvm/Target/TargetMachine.h b/include/llvm/Target/TargetMachine.h index bdcc4eff675..a8db68c5978 100644 --- a/include/llvm/Target/TargetMachine.h +++ b/include/llvm/Target/TargetMachine.h @@ -78,6 +78,14 @@ namespace CodeGenOpt { }; } +namespace FloatABI { + enum ABIType { + Default, + Soft, + Hard + }; +} + //===----------------------------------------------------------------------===// /// /// TargetMachine - Primary interface to the complete machine description for @@ -88,7 +96,7 @@ class TargetMachine { TargetMachine(const TargetMachine &); // DO NOT IMPLEMENT void operator=(const TargetMachine &); // DO NOT IMPLEMENT protected: // Can only create subclasses. - TargetMachine() : AsmInfo(0) { } + TargetMachine(); /// getSubtargetImpl - virtual method implemented by subclasses that returns /// a reference to that target's TargetSubtarget-derived member variable. diff --git a/include/llvm/Target/TargetOptions.h b/include/llvm/Target/TargetOptions.h index 0c74fa1f2c1..377e03f95c4 100644 --- a/include/llvm/Target/TargetOptions.h +++ b/include/llvm/Target/TargetOptions.h @@ -73,6 +73,14 @@ namespace llvm { /// target FP instructions. extern bool UseSoftFloat; + /// FloatABIType - This setting is set by -float-abi=xxx option is specfied + /// on the command line. This setting may either be Default, Soft, or Hard. + /// Default selects the target's default behavior. Soft selects the ABI for + /// UseSoftFloat, but does not inidcate that FP hardware may not be used. + /// Such a combination is unfortunately popular (e.g. arm-apple-darwin). + /// Hard presumes that the normal FP ABI is used. + extern FloatABI::ABIType FloatABIType; + /// NoZerosInBSS - By default some codegens place zero-initialized data to /// .bss section. This flag disables such behaviour (necessary, e.g. for /// crt*.o compiling). diff --git a/include/llvm/Target/TargetSelectionDAG.td b/include/llvm/Target/TargetSelectionDAG.td index 2cd29676dbf..364d4d0d3cc 100644 --- a/include/llvm/Target/TargetSelectionDAG.td +++ b/include/llvm/Target/TargetSelectionDAG.td @@ -228,6 +228,7 @@ class SDNodegetType()) < getTypeSizeInBits(Ty) && + "This is not an extending conversion!"); + assert(isSCEVable(Ty) && + "This is not a conversion to a SCEVable type!"); + Ty = getEffectiveSCEVType(Ty); + + // Sign-extend negative constants. + if (const SCEVConstant *SC = dyn_cast(Op)) + if (SC->getValue()->getValue().isNegative()) + return getSignExtendExpr(Op, Ty); + + // Peel off a truncate cast. + if (const SCEVTruncateExpr *T = dyn_cast(Op)) { + SCEVHandle NewOp = T->getOperand(); + if (getTypeSizeInBits(NewOp->getType()) < getTypeSizeInBits(Ty)) + return getAnyExtendExpr(NewOp, Ty); + return getTruncateOrNoop(NewOp, Ty); + } + + // Next try a zext cast. If the cast is folded, use it. + SCEVHandle ZExt = getZeroExtendExpr(Op, Ty); + if (!isa(ZExt)) + return ZExt; + + // Next try a sext cast. If the cast is folded, use it. + SCEVHandle SExt = getSignExtendExpr(Op, Ty); + if (!isa(SExt)) + return SExt; + + // If the expression is obviously signed, use the sext cast value. + if (isa(Op)) + return SExt; + + // Absent any other information, use the zext cast value. + return ZExt; +} + /// getAddExpr - Get a canonical add expression, or something simpler if /// possible. SCEVHandle ScalarEvolution::getAddExpr(std::vector &Ops) { @@ -1903,6 +1945,23 @@ ScalarEvolution::getNoopOrSignExtend(const SCEVHandle &V, const Type *Ty) { return getSignExtendExpr(V, Ty); } +/// getNoopOrAnyExtend - Return a SCEV corresponding to a conversion of +/// the input value to the specified type. If the type must be extended, +/// it is extended with unspecified bits. The conversion must not be +/// narrowing. +SCEVHandle +ScalarEvolution::getNoopOrAnyExtend(const SCEVHandle &V, const Type *Ty) { + const Type *SrcTy = V->getType(); + assert((SrcTy->isInteger() || (TD && isa(SrcTy))) && + (Ty->isInteger() || (TD && isa(Ty))) && + "Cannot noop or any extend with non-integer arguments!"); + assert(getTypeSizeInBits(SrcTy) <= getTypeSizeInBits(Ty) && + "getNoopOrAnyExtend cannot truncate!"); + if (getTypeSizeInBits(SrcTy) == getTypeSizeInBits(Ty)) + return V; // No conversion + return getAnyExtendExpr(V, Ty); +} + /// getTruncateOrNoop - Return a SCEV corresponding to a conversion of the /// input value to the specified type. The conversion must not be widening. SCEVHandle diff --git a/lib/Analysis/ScalarEvolutionExpander.cpp b/lib/Analysis/ScalarEvolutionExpander.cpp index ef77e46fc99..e1f8fa421f5 100644 --- a/lib/Analysis/ScalarEvolutionExpander.cpp +++ b/lib/Analysis/ScalarEvolutionExpander.cpp @@ -16,6 +16,7 @@ #include "llvm/Analysis/ScalarEvolutionExpander.h" #include "llvm/Analysis/LoopInfo.h" #include "llvm/Target/TargetData.h" +#include "llvm/ADT/STLExtras.h" using namespace llvm; /// InsertCastOfTo - Insert a cast of V to the specified type, doing what @@ -319,8 +320,7 @@ Value *SCEVExpander::expandAddToGEP(const SCEVHandle *op_begin, if (!AnyNonZeroIndices) { V = InsertNoopCastOfTo(V, Type::Int8Ty->getPointerTo(PTy->getAddressSpace())); - Value *Idx = expand(SE.getAddExpr(Ops)); - Idx = InsertNoopCastOfTo(Idx, Ty); + Value *Idx = expandCodeFor(SE.getAddExpr(Ops), Ty); // Fold a GEP with constant operands. if (Constant *CLHS = dyn_cast(V)) @@ -374,8 +374,7 @@ Value *SCEVExpander::visitAddExpr(const SCEVAddExpr *S) { // Emit a bunch of add instructions for (int i = S->getNumOperands()-2; i >= 0; --i) { - Value *W = expand(S->getOperand(i)); - W = InsertNoopCastOfTo(W, Ty); + Value *W = expandCodeFor(S->getOperand(i), Ty); V = InsertBinop(Instruction::Add, V, W, InsertPt); } return V; @@ -389,13 +388,11 @@ Value *SCEVExpander::visitMulExpr(const SCEVMulExpr *S) { FirstOp = 1; int i = S->getNumOperands()-2; - Value *V = expand(S->getOperand(i+1)); - V = InsertNoopCastOfTo(V, Ty); + Value *V = expandCodeFor(S->getOperand(i+1), Ty); // Emit a bunch of multiply instructions for (; i >= FirstOp; --i) { - Value *W = expand(S->getOperand(i)); - W = InsertNoopCastOfTo(W, Ty); + Value *W = expandCodeFor(S->getOperand(i), Ty); V = InsertBinop(Instruction::Mul, V, W, InsertPt); } @@ -408,8 +405,7 @@ Value *SCEVExpander::visitMulExpr(const SCEVMulExpr *S) { Value *SCEVExpander::visitUDivExpr(const SCEVUDivExpr *S) { const Type *Ty = SE.getEffectiveSCEVType(S->getType()); - Value *LHS = expand(S->getLHS()); - LHS = InsertNoopCastOfTo(LHS, Ty); + Value *LHS = expandCodeFor(S->getLHS(), Ty); if (const SCEVConstant *SC = dyn_cast(S->getRHS())) { const APInt &RHS = SC->getValue()->getValue(); if (RHS.isPowerOf2()) @@ -418,8 +414,7 @@ Value *SCEVExpander::visitUDivExpr(const SCEVUDivExpr *S) { InsertPt); } - Value *RHS = expand(S->getRHS()); - RHS = InsertNoopCastOfTo(RHS, Ty); + Value *RHS = expandCodeFor(S->getRHS(), Ty); return InsertBinop(Instruction::UDiv, LHS, RHS, InsertPt); } @@ -448,6 +443,34 @@ Value *SCEVExpander::visitAddRecExpr(const SCEVAddRecExpr *S) { const Type *Ty = SE.getEffectiveSCEVType(S->getType()); const Loop *L = S->getLoop(); + // First check for an existing canonical IV in a suitable type. + PHINode *CanonicalIV = 0; + if (PHINode *PN = L->getCanonicalInductionVariable()) + if (SE.isSCEVable(PN->getType()) && + isa(SE.getEffectiveSCEVType(PN->getType())) && + SE.getTypeSizeInBits(PN->getType()) >= SE.getTypeSizeInBits(Ty)) + CanonicalIV = PN; + + // Rewrite an AddRec in terms of the canonical induction variable, if + // its type is more narrow. + if (CanonicalIV && + SE.getTypeSizeInBits(CanonicalIV->getType()) > + SE.getTypeSizeInBits(Ty)) { + SCEVHandle Start = SE.getAnyExtendExpr(S->getStart(), + CanonicalIV->getType()); + SCEVHandle Step = SE.getAnyExtendExpr(S->getStepRecurrence(SE), + CanonicalIV->getType()); + Value *V = expand(SE.getAddRecExpr(Start, Step, S->getLoop())); + BasicBlock::iterator SaveInsertPt = getInsertionPoint(); + BasicBlock::iterator NewInsertPt = + next(BasicBlock::iterator(cast(V))); + while (isa(NewInsertPt)) ++NewInsertPt; + V = expandCodeFor(SE.getTruncateExpr(SE.getUnknown(V), Ty), 0, + NewInsertPt); + setInsertionPoint(SaveInsertPt); + return V; + } + // {X,+,F} --> X + {0,+,F} if (!S->getStart()->isZero()) { std::vector NewOps(S->getOperands()); @@ -481,6 +504,14 @@ Value *SCEVExpander::visitAddRecExpr(const SCEVAddRecExpr *S) { // {0,+,1} --> Insert a canonical induction variable into the loop! if (S->isAffine() && S->getOperand(1) == SE.getIntegerSCEV(1, Ty)) { + // If there's a canonical IV, just use it. + if (CanonicalIV) { + assert(Ty == SE.getEffectiveSCEVType(CanonicalIV->getType()) && + "IVs with types different from the canonical IV should " + "already have been handled!"); + return CanonicalIV; + } + // Create and insert the PHI node for the induction variable in the // specified loop. BasicBlock *Header = L->getHeader(); @@ -508,19 +539,16 @@ Value *SCEVExpander::visitAddRecExpr(const SCEVAddRecExpr *S) { return PN; } + // {0,+,F} --> {0,+,1} * F // Get the canonical induction variable I for this loop. - Value *I = getOrInsertCanonicalInductionVariable(L, Ty); + Value *I = CanonicalIV ? + CanonicalIV : + getOrInsertCanonicalInductionVariable(L, Ty); // If this is a simple linear addrec, emit it now as a special case. if (S->isAffine()) { // {0,+,F} --> i*F - Value *F = expand(S->getOperand(1)); - F = InsertNoopCastOfTo(F, Ty); - - // IF the step is by one, just return the inserted IV. - if (ConstantInt *CI = dyn_cast(F)) - if (CI->getValue() == 1) - return I; - + Value *F = expandCodeFor(S->getOperand(1), Ty); + // If the insert point is directly inside of the loop, emit the multiply at // the insert point. Otherwise, L is a loop that is a parent of the insert // point loop. If we can, move the multiply to the outer most loop that it @@ -555,16 +583,24 @@ Value *SCEVExpander::visitAddRecExpr(const SCEVAddRecExpr *S) { // into this folder. SCEVHandle IH = SE.getUnknown(I); // Get I as a "symbolic" SCEV. - SCEVHandle V = S->evaluateAtIteration(IH, SE); + // Promote S up to the canonical IV type, if the cast is foldable. + SCEVHandle NewS = S; + SCEVHandle Ext = SE.getNoopOrAnyExtend(S, I->getType()); + if (isa(Ext)) + NewS = Ext; + + SCEVHandle V = cast(NewS)->evaluateAtIteration(IH, SE); //cerr << "Evaluated: " << *this << "\n to: " << *V << "\n"; + // Truncate the result down to the original type, if needed. + SCEVHandle T = SE.getTruncateOrNoop(V, Ty); return expand(V); } Value *SCEVExpander::visitTruncateExpr(const SCEVTruncateExpr *S) { const Type *Ty = SE.getEffectiveSCEVType(S->getType()); - Value *V = expand(S->getOperand()); - V = InsertNoopCastOfTo(V, SE.getEffectiveSCEVType(V->getType())); + Value *V = expandCodeFor(S->getOperand(), + SE.getEffectiveSCEVType(S->getOperand()->getType())); Instruction *I = new TruncInst(V, Ty, "tmp.", InsertPt); InsertedValues.insert(I); return I; @@ -572,8 +608,8 @@ Value *SCEVExpander::visitTruncateExpr(const SCEVTruncateExpr *S) { Value *SCEVExpander::visitZeroExtendExpr(const SCEVZeroExtendExpr *S) { const Type *Ty = SE.getEffectiveSCEVType(S->getType()); - Value *V = expand(S->getOperand()); - V = InsertNoopCastOfTo(V, SE.getEffectiveSCEVType(V->getType())); + Value *V = expandCodeFor(S->getOperand(), + SE.getEffectiveSCEVType(S->getOperand()->getType())); Instruction *I = new ZExtInst(V, Ty, "tmp.", InsertPt); InsertedValues.insert(I); return I; @@ -581,8 +617,8 @@ Value *SCEVExpander::visitZeroExtendExpr(const SCEVZeroExtendExpr *S) { Value *SCEVExpander::visitSignExtendExpr(const SCEVSignExtendExpr *S) { const Type *Ty = SE.getEffectiveSCEVType(S->getType()); - Value *V = expand(S->getOperand()); - V = InsertNoopCastOfTo(V, SE.getEffectiveSCEVType(V->getType())); + Value *V = expandCodeFor(S->getOperand(), + SE.getEffectiveSCEVType(S->getOperand()->getType())); Instruction *I = new SExtInst(V, Ty, "tmp.", InsertPt); InsertedValues.insert(I); return I; @@ -590,11 +626,9 @@ Value *SCEVExpander::visitSignExtendExpr(const SCEVSignExtendExpr *S) { Value *SCEVExpander::visitSMaxExpr(const SCEVSMaxExpr *S) { const Type *Ty = SE.getEffectiveSCEVType(S->getType()); - Value *LHS = expand(S->getOperand(0)); - LHS = InsertNoopCastOfTo(LHS, Ty); + Value *LHS = expandCodeFor(S->getOperand(0), Ty); for (unsigned i = 1; i < S->getNumOperands(); ++i) { - Value *RHS = expand(S->getOperand(i)); - RHS = InsertNoopCastOfTo(RHS, Ty); + Value *RHS = expandCodeFor(S->getOperand(i), Ty); Instruction *ICmp = new ICmpInst(ICmpInst::ICMP_SGT, LHS, RHS, "tmp", InsertPt); InsertedValues.insert(ICmp); @@ -607,11 +641,9 @@ Value *SCEVExpander::visitSMaxExpr(const SCEVSMaxExpr *S) { Value *SCEVExpander::visitUMaxExpr(const SCEVUMaxExpr *S) { const Type *Ty = SE.getEffectiveSCEVType(S->getType()); - Value *LHS = expand(S->getOperand(0)); - LHS = InsertNoopCastOfTo(LHS, Ty); + Value *LHS = expandCodeFor(S->getOperand(0), Ty); for (unsigned i = 1; i < S->getNumOperands(); ++i) { - Value *RHS = expand(S->getOperand(i)); - RHS = InsertNoopCastOfTo(RHS, Ty); + Value *RHS = expandCodeFor(S->getOperand(i), Ty); Instruction *ICmp = new ICmpInst(ICmpInst::ICMP_UGT, LHS, RHS, "tmp", InsertPt); InsertedValues.insert(ICmp); diff --git a/lib/Archive/ArchiveWriter.cpp b/lib/Archive/ArchiveWriter.cpp index 336a2bdc658..cebb0877457 100644 --- a/lib/Archive/ArchiveWriter.cpp +++ b/lib/Archive/ArchiveWriter.cpp @@ -167,10 +167,11 @@ Archive::addFileBefore(const sys::Path& filePath, iterator where, mbr->data = 0; mbr->path = filePath; const sys::FileStatus *FSInfo = mbr->path.getFileStatus(false, ErrMsg); - if (FSInfo) - mbr->info = *FSInfo; - else + if (!FSInfo) { + delete mbr; return true; + } + mbr->info = *FSInfo; unsigned flags = 0; bool hasSlash = filePath.toString().find('/') != std::string::npos; diff --git a/lib/Bitcode/Writer/BitcodeWriter.cpp b/lib/Bitcode/Writer/BitcodeWriter.cpp index 9f16728d492..6dcddedef1c 100644 --- a/lib/Bitcode/Writer/BitcodeWriter.cpp +++ b/lib/Bitcode/Writer/BitcodeWriter.cpp @@ -1308,16 +1308,6 @@ static void WriteModule(const Module *M, BitstreamWriter &Stream) { // Emit constants. WriteModuleConstants(VE, Stream); - // If we have any aggregate values in the value table, purge them - these can - // only be used to initialize global variables. Doing so makes the value - // namespace smaller for code in functions. - int NumNonAggregates = VE.PurgeAggregateValues(); - if (NumNonAggregates != -1) { - SmallVector Vals; - Vals.push_back(NumNonAggregates); - Stream.EmitRecord(bitc::MODULE_CODE_PURGEVALS, Vals); - } - // Emit function bodies. for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I) if (!I->isDeclaration()) diff --git a/lib/Bitcode/Writer/ValueEnumerator.cpp b/lib/Bitcode/Writer/ValueEnumerator.cpp index 8002a36b474..32b2819762d 100644 --- a/lib/Bitcode/Writer/ValueEnumerator.cpp +++ b/lib/Bitcode/Writer/ValueEnumerator.cpp @@ -277,22 +277,6 @@ void ValueEnumerator::EnumerateAttributes(const AttrListPtr &PAL) { } -/// PurgeAggregateValues - If there are any aggregate values at the end of the -/// value list, remove them and return the count of the remaining values. If -/// there are none, return -1. -int ValueEnumerator::PurgeAggregateValues() { - // If there are no aggregate values at the end of the list, return -1. - if (Values.empty() || Values.back().first->getType()->isSingleValueType()) - return -1; - - // Otherwise, remove aggregate values... - while (!Values.empty() && !Values.back().first->getType()->isSingleValueType()) - Values.pop_back(); - - // ... and return the new size. - return Values.size(); -} - void ValueEnumerator::incorporateFunction(const Function &F) { NumModuleValues = Values.size(); diff --git a/lib/Bitcode/Writer/ValueEnumerator.h b/lib/Bitcode/Writer/ValueEnumerator.h index bb0324b1c57..40eeabb2b6a 100644 --- a/lib/Bitcode/Writer/ValueEnumerator.h +++ b/lib/Bitcode/Writer/ValueEnumerator.h @@ -99,11 +99,6 @@ public: return Attributes; } - /// PurgeAggregateValues - If there are any aggregate values at the end of the - /// value list, remove them and return the count of the remaining values. If - /// there are none, return -1. - int PurgeAggregateValues(); - /// incorporateFunction/purgeFunction - If you'd like to deal with a function, /// use these two methods to get its data into the ValueEnumerator! /// diff --git a/lib/CodeGen/AsmPrinter/DwarfDebug.cpp b/lib/CodeGen/AsmPrinter/DwarfDebug.cpp index 5a66f4b04b9..c773378bb01 100644 --- a/lib/CodeGen/AsmPrinter/DwarfDebug.cpp +++ b/lib/CodeGen/AsmPrinter/DwarfDebug.cpp @@ -1581,6 +1581,7 @@ void DwarfDebug::EndFunction(MachineFunction *MF) { FunctionDbgScope = NULL; LexicalScopeStack.clear(); AbstractInstanceRootList.clear(); + AbstractInstanceRootMap.clear(); } Lines.clear(); @@ -1669,7 +1670,11 @@ unsigned DwarfDebug::RecordRegionEnd(GlobalVariable *V) { DbgScope *Scope = getOrCreateScope(V); unsigned ID = MMI->NextLabelID(); Scope->setEndLabelID(ID); - if (LexicalScopeStack.size() != 0) + // FIXME : region.end() may not be in the last basic block. + // For now, do not pop last lexical scope because next basic + // block may start new inlined function's body. + unsigned LSSize = LexicalScopeStack.size(); + if (LSSize != 0 && LSSize != 1) LexicalScopeStack.pop_back(); if (TimePassesIsEnabled) diff --git a/lib/CodeGen/CMakeLists.txt b/lib/CodeGen/CMakeLists.txt index ff917a72f0c..5ba8b3c147a 100644 --- a/lib/CodeGen/CMakeLists.txt +++ b/lib/CodeGen/CMakeLists.txt @@ -12,6 +12,7 @@ add_llvm_library(LLVMCodeGen IntrinsicLowering.cpp LLVMTargetMachine.cpp LatencyPriorityQueue.cpp + LazyLiveness.cpp LiveInterval.cpp LiveIntervalAnalysis.cpp LiveStackAnalysis.cpp diff --git a/lib/CodeGen/ELF.h b/lib/CodeGen/ELF.h index c22f6ed92e7..796bc2c3b2c 100644 --- a/lib/CodeGen/ELF.h +++ b/lib/CodeGen/ELF.h @@ -10,23 +10,24 @@ // This header contains common, non-processor-specific data structures and // constants for the ELF file format. // -// The details of the ELF32 bits in this file are largely based on -// the Tool Interface Standard (TIS) Executable and Linking Format -// (ELF) Specification Version 1.2, May 1995. The ELF64 stuff is not -// standardized, as far as I can tell. It was largely based on information -// I found in OpenBSD header files. +// The details of the ELF32 bits in this file are largely based on the Tool +// Interface Standard (TIS) Executable and Linking Format (ELF) Specification +// Version 1.2, May 1995. The ELF64 is based on HP/Intel definition of the +// ELF-64 object file format document, Version 1.5 Draft 2 May 27, 1998 // //===----------------------------------------------------------------------===// #ifndef CODEGEN_ELF_H #define CODEGEN_ELF_H +#include "llvm/GlobalVariable.h" +#include "llvm/CodeGen/BinaryObject.h" #include "llvm/CodeGen/MachineRelocation.h" #include "llvm/Support/DataTypes.h" #include namespace llvm { - class GlobalVariable; + class BinaryObject; // Identification Indexes enum { @@ -47,71 +48,28 @@ namespace llvm { ET_HIPROC = 0xffff // Processor-specific }; - // Object file classes. - enum { - ELFCLASS32 = 1, // 32-bit object file - ELFCLASS64 = 2 // 64-bit object file - }; - - // Object file byte orderings. - enum { - ELFDATA2LSB = 1, // Little-endian object file - ELFDATA2MSB = 2 // Big-endian object file - }; - // Versioning enum { EV_NONE = 0, EV_CURRENT = 1 }; - struct ELFHeader { - // e_machine - This field is the target specific value to emit as the - // e_machine member of the ELF header. - unsigned short e_machine; - - // e_flags - The machine flags for the target. This defaults to zero. - unsigned e_flags; - - // e_size - Holds the ELF header's size in bytes - unsigned e_ehsize; - - // Endianess and ELF Class (64 or 32 bits) - unsigned ByteOrder; - unsigned ElfClass; - - unsigned getByteOrder() const { return ByteOrder; } - unsigned getElfClass() const { return ElfClass; } - unsigned getSize() const { return e_ehsize; } - unsigned getMachine() const { return e_machine; } - unsigned getFlags() const { return e_flags; } - - ELFHeader(unsigned short machine, unsigned flags, - bool is64Bit, bool isLittleEndian) - : e_machine(machine), e_flags(flags) { - ElfClass = is64Bit ? ELFCLASS64 : ELFCLASS32; - ByteOrder = isLittleEndian ? ELFDATA2LSB : ELFDATA2MSB; - e_ehsize = is64Bit ? 64 : 52; - } - }; - /// ELFSection - This struct contains information about each section that is /// emitted to the file. This is eventually turned into the section header /// table at the end of the file. - struct ELFSection { - + class ELFSection : public BinaryObject { + public: // ELF specific fields - std::string Name; // Name of the section. - unsigned NameIdx; // Index in .shstrtab of name, once emitted. - unsigned Type; - unsigned Flags; - uint64_t Addr; - unsigned Offset; - unsigned Size; - unsigned Link; - unsigned Info; - unsigned Align; - unsigned EntSize; + unsigned NameIdx; // sh_name - .shstrtab idx of name, once emitted. + unsigned Type; // sh_type - Section contents & semantics + unsigned Flags; // sh_flags - Section flags. + uint64_t Addr; // sh_addr - The mem addr this section is in. + unsigned Offset; // sh_offset - Offset from the file start + unsigned Size; // sh_size - The section size. + unsigned Link; // sh_link - Section header table index link. + unsigned Info; // sh_info - Auxillary information. + unsigned Align; // sh_addralign - Alignment of section. + unsigned EntSize; // sh_entsize - Size of entries in the section e // Section Header Flags enum { @@ -141,8 +99,8 @@ namespace llvm { SHT_REL = 9, // Relocation entries; no explicit addends. SHT_SHLIB = 10, // Reserved. SHT_DYNSYM = 11, // Symbol table. - SHT_LOPROC = 0x70000000, // Lowest processor architecture-specific type. - SHT_HIPROC = 0x7fffffff, // Highest processor architecture-specific type. + SHT_LOPROC = 0x70000000, // Lowest processor arch-specific type. + SHT_HIPROC = 0x7fffffff, // Highest processor arch-specific type. SHT_LOUSER = 0x80000000, // Lowest type reserved for applications. SHT_HIUSER = 0xffffffff // Highest type reserved for applications. }; @@ -161,22 +119,9 @@ namespace llvm { /// SectionIdx - The number of the section in the Section Table. unsigned short SectionIdx; - /// SectionData - The actual data for this section which we are building - /// up for emission to the file. - std::vector SectionData; - - /// Relocations - The relocations that we have encountered so far in this - /// section that we will need to convert to Elf relocation entries when - /// the file is written. - std::vector Relocations; - - /// Section Header Size - static unsigned getSectionHdrSize(bool is64Bit) - { return is64Bit ? 64 : 40; } - - ELFSection(const std::string &name) - : Name(name), Type(0), Flags(0), Addr(0), Offset(0), Size(0), - Link(0), Info(0), Align(0), EntSize(0) {} + ELFSection(const std::string &name, bool isLittleEndian, bool is64Bit) + : BinaryObject(name, isLittleEndian, is64Bit), Type(0), Flags(0), Addr(0), + Offset(0), Size(0), Link(0), Info(0), Align(0), EntSize(0) {} }; /// ELFSym - This struct contains information about each symbol that is @@ -207,9 +152,33 @@ namespace llvm { STT_FILE = 4 }; + enum { + STV_DEFAULT = 0, // Visibility is specified by binding type + STV_INTERNAL = 1, // Defined by processor supplements + STV_HIDDEN = 2, // Not visible to other components + STV_PROTECTED = 3 // Visible in other components but not preemptable + }; + ELFSym(const GlobalValue *gv) : GV(gv), NameIdx(0), Value(0), Size(0), Info(0), Other(0), - SectionIdx(ELFSection::SHN_UNDEF) {} + SectionIdx(ELFSection::SHN_UNDEF) { + if (!GV) + return; + + switch (GV->getVisibility()) { + default: + assert(0 && "unknown visibility type"); + case GlobalValue::DefaultVisibility: + Other = STV_DEFAULT; + break; + case GlobalValue::HiddenVisibility: + Other = STV_HIDDEN; + break; + case GlobalValue::ProtectedVisibility: + Other = STV_PROTECTED; + break; + } + } void SetBind(unsigned X) { assert(X == (X & 0xF) && "Bind value out of range!"); diff --git a/lib/CodeGen/ELFCodeEmitter.cpp b/lib/CodeGen/ELFCodeEmitter.cpp index c7bd8736bb9..ca683969e41 100644 --- a/lib/CodeGen/ELFCodeEmitter.cpp +++ b/lib/CodeGen/ELFCodeEmitter.cpp @@ -13,9 +13,9 @@ #include "llvm/Constants.h" #include "llvm/DerivedTypes.h" #include "llvm/Function.h" +#include "llvm/CodeGen/BinaryObject.h" #include "llvm/CodeGen/MachineConstantPool.h" #include "llvm/CodeGen/MachineJumpTableInfo.h" -#include "llvm/Target/TargetData.h" #include "llvm/Target/TargetMachine.h" #include "llvm/Support/Debug.h" @@ -28,27 +28,22 @@ namespace llvm { /// startFunction - This callback is invoked when a new machine function is /// about to be emitted. void ELFCodeEmitter::startFunction(MachineFunction &MF) { - const TargetData *TD = TM.getTargetData(); - const Function *F = MF.getFunction(); - - // Align the output buffer to the appropriate alignment, power of 2. - unsigned FnAlign = F->getAlignment(); - unsigned TDAlign = TD->getPrefTypeAlignment(F->getType()); - unsigned Align = std::max(FnAlign, TDAlign); - assert(!(Align & (Align-1)) && "Alignment is not a power of two!"); - // Get the ELF Section that this function belongs in. ES = &EW.getTextSection(); + DOUT << "processing function: " << MF.getFunction()->getName() << "\n"; + // FIXME: better memory management, this will be replaced by BinaryObjects - ES->SectionData.reserve(4096); - BufferBegin = &ES->SectionData[0]; - BufferEnd = BufferBegin + ES->SectionData.capacity(); + BinaryData &BD = ES->getData(); + BD.reserve(4096); + BufferBegin = &BD[0]; + BufferEnd = BufferBegin + BD.capacity(); - // Upgrade the section alignment if required. + // Align the output buffer with function alignment, and + // upgrade the section alignment if required + unsigned Align = + TM.getELFWriterInfo()->getFunctionAlignment(MF.getFunction()); if (ES->Align < Align) ES->Align = Align; - - // Round the size up to the correct alignment for starting the new function. ES->Size = (ES->Size + (Align-1)) & (-Align); // Snaity check on allocated space for text section @@ -107,7 +102,7 @@ bool ELFCodeEmitter::finishFunction(MachineFunction &MF) { FnSym.Value = FnStartPtr-BufferBegin; // Finally, add it to the symtab. - EW.SymbolTable.push_back(FnSym); + EW.SymbolList.push_back(FnSym); // Relocations // ----------- @@ -128,7 +123,7 @@ bool ELFCodeEmitter::finishFunction(MachineFunction &MF) { } else { assert(0 && "Unhandled relocation type"); } - ES->Relocations.push_back(MR); + ES->addRelocation(MR); } Relocations.clear(); diff --git a/lib/CodeGen/ELFWriter.cpp b/lib/CodeGen/ELFWriter.cpp index 3859ea3865c..aeccefbd3e8 100644 --- a/lib/CodeGen/ELFWriter.cpp +++ b/lib/CodeGen/ELFWriter.cpp @@ -26,9 +26,6 @@ // ... // #N. ".shstrtab" entry - String table for the section names. // -// NOTE: This code should eventually be extended to support 64-bit ELF (this -// won't be hard), but we haven't done so yet! -// //===----------------------------------------------------------------------===// #define DEBUG_TYPE "elfwriter" @@ -36,18 +33,18 @@ #include "ELFWriter.h" #include "ELFCodeEmitter.h" #include "ELF.h" +#include "llvm/Constants.h" #include "llvm/Module.h" #include "llvm/PassManager.h" #include "llvm/DerivedTypes.h" +#include "llvm/CodeGen/BinaryObject.h" #include "llvm/CodeGen/FileWriters.h" #include "llvm/CodeGen/MachineCodeEmitter.h" #include "llvm/CodeGen/MachineConstantPool.h" #include "llvm/CodeGen/MachineFunctionPass.h" #include "llvm/Target/TargetData.h" -#include "llvm/Target/TargetELFWriterInfo.h" #include "llvm/Target/TargetMachine.h" #include "llvm/Support/Mangler.h" -#include "llvm/Support/OutputBuffer.h" #include "llvm/Support/Streams.h" #include "llvm/Support/raw_ostream.h" #include "llvm/Support/Debug.h" @@ -70,21 +67,23 @@ MachineCodeEmitter *llvm::AddELFWriter(PassManagerBase &PM, //===----------------------------------------------------------------------===// ELFWriter::ELFWriter(raw_ostream &o, TargetMachine &tm) - : MachineFunctionPass(&ID), O(o), TM(tm), ElfHdr() { - is64Bit = TM.getTargetData()->getPointerSizeInBits() == 64; - isLittleEndian = TM.getTargetData()->isLittleEndian(); + : MachineFunctionPass(&ID), O(o), TM(tm), + is64Bit(TM.getTargetData()->getPointerSizeInBits() == 64), + isLittleEndian(TM.getTargetData()->isLittleEndian()), + ElfHdr(isLittleEndian, is64Bit) { - ElfHdr = new ELFHeader(TM.getELFWriterInfo()->getEMachine(), 0, - is64Bit, isLittleEndian); + TAI = TM.getTargetAsmInfo(); + TEW = TM.getELFWriterInfo(); // Create the machine code emitter object for this target. MCE = new ELFCodeEmitter(*this); + + // Inital number of sections NumSections = 0; } ELFWriter::~ELFWriter() { delete MCE; - delete ElfHdr; } // doInitialization - Emit the file header and all of the global variables for @@ -92,10 +91,6 @@ ELFWriter::~ELFWriter() { bool ELFWriter::doInitialization(Module &M) { Mang = new Mangler(M); - // Local alias to shortenify coming code. - std::vector &FH = FileHeader; - OutputBuffer FHOut(FH, is64Bit, isLittleEndian); - // ELF Header // ---------- // Fields e_shnum e_shstrndx are only known after all section have @@ -104,54 +99,58 @@ bool ELFWriter::doInitialization(Module &M) { // // Note // ---- - // FHOut.outaddr method behaves differently for ELF32 and ELF64 writing + // emitWord method behaves differently for ELF32 and ELF64, writing // 4 bytes in the former and 8 in the last for *_off and *_addr elf types - FHOut.outbyte(0x7f); // e_ident[EI_MAG0] - FHOut.outbyte('E'); // e_ident[EI_MAG1] - FHOut.outbyte('L'); // e_ident[EI_MAG2] - FHOut.outbyte('F'); // e_ident[EI_MAG3] + ElfHdr.emitByte(0x7f); // e_ident[EI_MAG0] + ElfHdr.emitByte('E'); // e_ident[EI_MAG1] + ElfHdr.emitByte('L'); // e_ident[EI_MAG2] + ElfHdr.emitByte('F'); // e_ident[EI_MAG3] - FHOut.outbyte(ElfHdr->getElfClass()); // e_ident[EI_CLASS] - FHOut.outbyte(ElfHdr->getByteOrder()); // e_ident[EI_DATA] - FHOut.outbyte(EV_CURRENT); // e_ident[EI_VERSION] + ElfHdr.emitByte(TEW->getEIClass()); // e_ident[EI_CLASS] + ElfHdr.emitByte(TEW->getEIData()); // e_ident[EI_DATA] + ElfHdr.emitByte(EV_CURRENT); // e_ident[EI_VERSION] + ElfHdr.emitAlignment(16); // e_ident[EI_NIDENT-EI_PAD] - FH.resize(16); // e_ident[EI_NIDENT-EI_PAD] - - FHOut.outhalf(ET_REL); // e_type - FHOut.outhalf(ElfHdr->getMachine()); // e_machine = target - FHOut.outword(EV_CURRENT); // e_version - FHOut.outaddr(0); // e_entry = 0, no entry point in .o file - FHOut.outaddr(0); // e_phoff = 0, no program header for .o - ELFHdr_e_shoff_Offset = FH.size(); - FHOut.outaddr(0); // e_shoff = sec hdr table off in bytes - FHOut.outword(ElfHdr->getFlags()); // e_flags = whatever the target wants - FHOut.outhalf(ElfHdr->getSize()); // e_ehsize = ELF header size - FHOut.outhalf(0); // e_phentsize = prog header entry size - FHOut.outhalf(0); // e_phnum = # prog header entries = 0 + ElfHdr.emitWord16(ET_REL); // e_type + ElfHdr.emitWord16(TEW->getEMachine()); // e_machine = target + ElfHdr.emitWord32(EV_CURRENT); // e_version + ElfHdr.emitWord(0); // e_entry, no entry point in .o file + ElfHdr.emitWord(0); // e_phoff, no program header for .o + ELFHdr_e_shoff_Offset = ElfHdr.size(); + ElfHdr.emitWord(0); // e_shoff = sec hdr table off in bytes + ElfHdr.emitWord32(TEW->getEFlags()); // e_flags = whatever the target wants + ElfHdr.emitWord16(TEW->getHdrSize()); // e_ehsize = ELF header size + ElfHdr.emitWord16(0); // e_phentsize = prog header entry size + ElfHdr.emitWord16(0); // e_phnum = # prog header entries = 0 // e_shentsize = Section header entry size - FHOut.outhalf(ELFSection::getSectionHdrSize(is64Bit)); + ElfHdr.emitWord16(TEW->getSHdrSize()); // e_shnum = # of section header ents - ELFHdr_e_shnum_Offset = FH.size(); - FHOut.outhalf(0); + ELFHdr_e_shnum_Offset = ElfHdr.size(); + ElfHdr.emitWord16(0); // Placeholder // e_shstrndx = Section # of '.shstrtab' - ELFHdr_e_shstrndx_Offset = FH.size(); - FHOut.outhalf(0); + ELFHdr_e_shstrndx_Offset = ElfHdr.size(); + ElfHdr.emitWord16(0); // Placeholder // Add the null section, which is required to be first in the file. getSection("", ELFSection::SHT_NULL, 0); - // Start up the symbol table. The first entry in the symtab is the null + // Start up the symbol table. The first entry in the symtab is the null // entry. - SymbolTable.push_back(ELFSym(0)); + SymbolList.push_back(ELFSym(0)); return false; } void ELFWriter::EmitGlobal(GlobalVariable *GV) { + + // XXX: put local symbols *before* global ones! + const Section *S = TAI->SectionForGlobal(GV); + DOUT << "Section " << S->getName() << " for global " << GV->getName() << "\n"; + // If this is an external global, emit it now. TODO: Note that it would be // better to ignore the symbol here and only add it to the symbol table if // referenced. @@ -160,17 +159,17 @@ void ELFWriter::EmitGlobal(GlobalVariable *GV) { ExternalSym.SetBind(ELFSym::STB_GLOBAL); ExternalSym.SetType(ELFSym::STT_NOTYPE); ExternalSym.SectionIdx = ELFSection::SHN_UNDEF; - SymbolTable.push_back(ExternalSym); + SymbolList.push_back(ExternalSym); return; } - unsigned Align = TM.getTargetData()->getPreferredAlignment(GV); - unsigned Size = - TM.getTargetData()->getTypeAllocSize(GV->getType()->getElementType()); + const TargetData *TD = TM.getTargetData(); + unsigned Align = TD->getPreferredAlignment(GV); + Constant *CV = GV->getInitializer(); + unsigned Size = TD->getTypeAllocSize(CV->getType()); - // If this global has a zero initializer, it is part of the .bss or common - // section. - if (GV->getInitializer()->isNullValue()) { + // If this global has a zero initializer, go to .bss or common section. + if (CV->isNullValue() || isa(CV)) { // If this global is part of the common block, add it now. Variables are // part of the common block if they are zero initialized and allowed to be // merged with other symbols. @@ -182,14 +181,14 @@ void ELFWriter::EmitGlobal(GlobalVariable *GV) { CommonSym.Size = Size; CommonSym.SetBind(ELFSym::STB_GLOBAL); CommonSym.SetType(ELFSym::STT_OBJECT); - // TODO SOMEDAY: add ELF visibility. CommonSym.SectionIdx = ELFSection::SHN_COMMON; - SymbolTable.push_back(CommonSym); + SymbolList.push_back(CommonSym); + getSection(S->getName(), ELFSection::SHT_NOBITS, + ELFSection::SHF_WRITE | ELFSection::SHF_ALLOC, 1); return; } // Otherwise, this symbol is part of the .bss section. Emit it now. - // Handle alignment. Ensure section is aligned at least as much as required // by this symbol. ELFSection &BSSSection = getBSSSection(); @@ -220,18 +219,128 @@ void ELFWriter::EmitGlobal(GlobalVariable *GV) { // Set the idx of the .bss section BSSSym.SectionIdx = BSSSection.SectionIdx; if (!GV->hasPrivateLinkage()) - SymbolTable.push_back(BSSSym); + SymbolList.push_back(BSSSym); // Reserve space in the .bss section for this symbol. BSSSection.Size += Size; return; } - // FIXME: handle .rodata - //assert(!GV->isConstant() && "unimp"); + /// Emit the Global symbol to the right ELF section + ELFSym GblSym(GV); + GblSym.Size = Size; + GblSym.SetType(ELFSym::STT_OBJECT); + GblSym.SetBind(ELFSym::STB_GLOBAL); + unsigned Flags = S->getFlags(); + unsigned SectType = ELFSection::SHT_PROGBITS; + unsigned SHdrFlags = ELFSection::SHF_ALLOC; - // FIXME: handle .data - //assert(0 && "unimp"); + if (Flags & SectionFlags::Code) + SHdrFlags |= ELFSection::SHF_EXECINSTR; + if (Flags & SectionFlags::Writeable) + SHdrFlags |= ELFSection::SHF_WRITE; + if (Flags & SectionFlags::Mergeable) + SHdrFlags |= ELFSection::SHF_MERGE; + if (Flags & SectionFlags::TLS) + SHdrFlags |= ELFSection::SHF_TLS; + if (Flags & SectionFlags::Strings) + SHdrFlags |= ELFSection::SHF_STRINGS; + + // Remove tab from section name prefix + std::string SectionName(S->getName()); + size_t Pos = SectionName.find("\t"); + if (Pos != std::string::npos) + SectionName.erase(Pos, 1); + + // The section alignment should be bound to the element with + // the largest alignment + ELFSection &ElfS = getSection(SectionName, SectType, SHdrFlags); + GblSym.SectionIdx = ElfS.SectionIdx; + if (Align > ElfS.Align) + ElfS.Align = Align; + + // S.Value should contain the symbol index inside the section, + // and all symbols should start on their required alignment boundary + GblSym.Value = (ElfS.size() + (Align-1)) & (-Align); + ElfS.emitAlignment(Align); + + // Emit the constant symbol to its section + EmitGlobalConstant(CV, ElfS); + SymbolList.push_back(GblSym); +} + +void ELFWriter::EmitGlobalConstantStruct(const ConstantStruct *CVS, + ELFSection &GblS) { + + // Print the fields in successive locations. Pad to align if needed! + const TargetData *TD = TM.getTargetData(); + unsigned Size = TD->getTypeAllocSize(CVS->getType()); + const StructLayout *cvsLayout = TD->getStructLayout(CVS->getType()); + uint64_t sizeSoFar = 0; + for (unsigned i = 0, e = CVS->getNumOperands(); i != e; ++i) { + const Constant* field = CVS->getOperand(i); + + // Check if padding is needed and insert one or more 0s. + uint64_t fieldSize = TD->getTypeAllocSize(field->getType()); + uint64_t padSize = ((i == e-1 ? Size : cvsLayout->getElementOffset(i+1)) + - cvsLayout->getElementOffset(i)) - fieldSize; + sizeSoFar += fieldSize + padSize; + + // Now print the actual field value. + EmitGlobalConstant(field, GblS); + + // Insert padding - this may include padding to increase the size of the + // current field up to the ABI size (if the struct is not packed) as well + // as padding to ensure that the next field starts at the right offset. + for (unsigned p=0; p < padSize; p++) + GblS.emitByte(0); + } + assert(sizeSoFar == cvsLayout->getSizeInBytes() && + "Layout of constant struct may be incorrect!"); +} + +void ELFWriter::EmitGlobalConstant(const Constant *CV, ELFSection &GblS) { + const TargetData *TD = TM.getTargetData(); + unsigned Size = TD->getTypeAllocSize(CV->getType()); + + if (const ConstantArray *CVA = dyn_cast(CV)) { + if (CVA->isString()) { + std::string GblStr = CVA->getAsString(); + GblS.emitString(GblStr); + } else { // Not a string. Print the values in successive locations + for (unsigned i = 0, e = CVA->getNumOperands(); i != e; ++i) + EmitGlobalConstant(CVA->getOperand(i), GblS); + } + return; + } else if (const ConstantStruct *CVS = dyn_cast(CV)) { + EmitGlobalConstantStruct(CVS, GblS); + return; + } else if (const ConstantFP *CFP = dyn_cast(CV)) { + uint64_t Val = CFP->getValueAPF().bitcastToAPInt().getZExtValue(); + if (CFP->getType() == Type::DoubleTy) + GblS.emitWord64(Val); + else if (CFP->getType() == Type::FloatTy) + GblS.emitWord32(Val); + else if (CFP->getType() == Type::X86_FP80Ty) { + assert(0 && "X86_FP80Ty global emission not implemented"); + } else if (CFP->getType() == Type::PPC_FP128Ty) + assert(0 && "PPC_FP128Ty global emission not implemented"); + return; + } else if (const ConstantInt *CI = dyn_cast(CV)) { + if (Size == 4) + GblS.emitWord32(CI->getZExtValue()); + else if (Size == 8) + GblS.emitWord64(CI->getZExtValue()); + else + assert(0 && "LargeInt global emission not implemented"); + return; + } else if (const ConstantVector *CP = dyn_cast(CV)) { + const VectorType *PTy = CP->getType(); + for (unsigned I = 0, E = PTy->getNumElements(); I < E; ++I) + EmitGlobalConstant(CP->getOperand(I), GblS); + return; + } + assert(0 && "unknown global constant"); } @@ -243,22 +352,41 @@ bool ELFWriter::runOnMachineFunction(MachineFunction &MF) { /// doFinalization - Now that the module has been completely processed, emit /// the ELF file to 'O'. bool ELFWriter::doFinalization(Module &M) { - // Okay, the ELF header and .text sections have been completed, build the - // .data, .bss, and "common" sections next. + /// FIXME: This should be removed when moving to ObjectCodeEmiter. Since the + /// current ELFCodeEmiter uses CurrBuff, ... it doesn't update S.Data + /// vector size for .text sections, so this is a quick dirty fix + ELFSection &TS = getTextSection(); + if (TS.Size) { + BinaryData &BD = TS.getData(); + for (unsigned e=0; egetNonexecutableStackDirective()) + getNonExecStackSection(); + // Emit the symbol table now, if non-empty. EmitSymbolTable(); // Emit the relocation sections. EmitRelocations(); - // Emit the string table for the sections in the ELF file. + // Emit the sections string table. EmitSectionTableStringTable(); - // Emit the sections to the .o file, and emit the section table for the file. + // Dump the sections and section table to the .o file. OutputSectionsAndSectionTable(); // We are done with the abstract symbols. @@ -274,78 +402,97 @@ bool ELFWriter::doFinalization(Module &M) { void ELFWriter::EmitRelocations() { } +/// EmitSymbol - Write symbol 'Sym' to the symbol table 'SymbolTable' +void ELFWriter::EmitSymbol(BinaryObject &SymbolTable, ELFSym &Sym) { + if (is64Bit) { + SymbolTable.emitWord32(Sym.NameIdx); + SymbolTable.emitByte(Sym.Info); + SymbolTable.emitByte(Sym.Other); + SymbolTable.emitWord16(Sym.SectionIdx); + SymbolTable.emitWord64(Sym.Value); + SymbolTable.emitWord64(Sym.Size); + } else { + SymbolTable.emitWord32(Sym.NameIdx); + SymbolTable.emitWord32(Sym.Value); + SymbolTable.emitWord32(Sym.Size); + SymbolTable.emitByte(Sym.Info); + SymbolTable.emitByte(Sym.Other); + SymbolTable.emitWord16(Sym.SectionIdx); + } +} + +/// EmitSectionHeader - Write section 'Section' header in 'SHdrTab' +/// Section Header Table +void ELFWriter::EmitSectionHeader(BinaryObject &SHdrTab, + const ELFSection &SHdr) { + SHdrTab.emitWord32(SHdr.NameIdx); + SHdrTab.emitWord32(SHdr.Type); + if (is64Bit) { + SHdrTab.emitWord64(SHdr.Flags); + SHdrTab.emitWord(SHdr.Addr); + SHdrTab.emitWord(SHdr.Offset); + SHdrTab.emitWord64(SHdr.Size); + SHdrTab.emitWord32(SHdr.Link); + SHdrTab.emitWord32(SHdr.Info); + SHdrTab.emitWord64(SHdr.Align); + SHdrTab.emitWord64(SHdr.EntSize); + } else { + SHdrTab.emitWord32(SHdr.Flags); + SHdrTab.emitWord(SHdr.Addr); + SHdrTab.emitWord(SHdr.Offset); + SHdrTab.emitWord32(SHdr.Size); + SHdrTab.emitWord32(SHdr.Link); + SHdrTab.emitWord32(SHdr.Info); + SHdrTab.emitWord32(SHdr.Align); + SHdrTab.emitWord32(SHdr.EntSize); + } +} + /// EmitSymbolTable - If the current symbol table is non-empty, emit the string /// table for it and then the symbol table itself. void ELFWriter::EmitSymbolTable() { - if (SymbolTable.size() == 1) return; // Only the null entry. + if (SymbolList.size() == 1) return; // Only the null entry. // FIXME: compact all local symbols to the start of the symtab. unsigned FirstNonLocalSymbol = 1; - ELFSection &StrTab = getSection(".strtab", ELFSection::SHT_STRTAB, 0); - StrTab.Align = 1; - - DataBuffer &StrTabBuf = StrTab.SectionData; - OutputBuffer StrTabOut(StrTabBuf, is64Bit, isLittleEndian); + ELFSection &StrTab = getStringTableSection(); // Set the zero'th symbol to a null byte, as required. - StrTabOut.outbyte(0); + StrTab.emitByte(0); + unsigned Index = 1; - for (unsigned i = 1, e = SymbolTable.size(); i != e; ++i) { + for (unsigned i = 1, e = SymbolList.size(); i != e; ++i) { // Use the name mangler to uniquify the LLVM symbol. - std::string Name = Mang->getValueName(SymbolTable[i].GV); + std::string Name = Mang->getValueName(SymbolList[i].GV); if (Name.empty()) { - SymbolTable[i].NameIdx = 0; + SymbolList[i].NameIdx = 0; } else { - SymbolTable[i].NameIdx = Index; - - // Add the name to the output buffer, including the null terminator. - StrTabBuf.insert(StrTabBuf.end(), Name.begin(), Name.end()); - - // Add a null terminator. - StrTabBuf.push_back(0); + SymbolList[i].NameIdx = Index; + StrTab.emitString(Name); // Keep track of the number of bytes emitted to this section. Index += Name.size()+1; } } - assert(Index == StrTabBuf.size()); + assert(Index == StrTab.size()); StrTab.Size = Index; // Now that we have emitted the string table and know the offset into the // string table of each symbol, emit the symbol table itself. - ELFSection &SymTab = getSection(".symtab", ELFSection::SHT_SYMTAB, 0); - SymTab.Align = is64Bit ? 8 : 4; - SymTab.Link = StrTab.SectionIdx; // Section Index of .strtab. - SymTab.Info = FirstNonLocalSymbol; // First non-STB_LOCAL symbol. - SymTab.EntSize = is64Bit ? 24 : 16; // Size of each symtab entry. - DataBuffer &SymTabBuf = SymTab.SectionData; - OutputBuffer SymTabOut(SymTabBuf, is64Bit, isLittleEndian); + ELFSection &SymTab = getSymbolTableSection(); + SymTab.Align = TEW->getSymTabAlignment(); + SymTab.Link = StrTab.SectionIdx; // Section Index of .strtab. + SymTab.Info = FirstNonLocalSymbol; // First non-STB_LOCAL symbol. - if (!is64Bit) { // 32-bit and 64-bit formats are shuffled a bit. - for (unsigned i = 0, e = SymbolTable.size(); i != e; ++i) { - ELFSym &Sym = SymbolTable[i]; - SymTabOut.outword(Sym.NameIdx); - SymTabOut.outaddr32(Sym.Value); - SymTabOut.outword(Sym.Size); - SymTabOut.outbyte(Sym.Info); - SymTabOut.outbyte(Sym.Other); - SymTabOut.outhalf(Sym.SectionIdx); - } - } else { - for (unsigned i = 0, e = SymbolTable.size(); i != e; ++i) { - ELFSym &Sym = SymbolTable[i]; - SymTabOut.outword(Sym.NameIdx); - SymTabOut.outbyte(Sym.Info); - SymTabOut.outbyte(Sym.Other); - SymTabOut.outhalf(Sym.SectionIdx); - SymTabOut.outaddr64(Sym.Value); - SymTabOut.outxword(Sym.Size); - } - } + // Size of each symtab entry. + SymTab.EntSize = TEW->getSymTabEntrySize(); - SymTab.Size = SymTabBuf.size(); + for (unsigned i = 0, e = SymbolList.size(); i != e; ++i) + EmitSymbol(SymTab, SymbolList[i]); + + SymTab.Size = SymTab.size(); } /// EmitSectionTableStringTable - This method adds and emits a section for the @@ -357,32 +504,25 @@ void ELFWriter::EmitSectionTableStringTable() { // Now that we know which section number is the .shstrtab section, update the // e_shstrndx entry in the ELF header. - OutputBuffer FHOut(FileHeader, is64Bit, isLittleEndian); - FHOut.fixhalf(SHStrTab.SectionIdx, ELFHdr_e_shstrndx_Offset); + ElfHdr.fixWord16(SHStrTab.SectionIdx, ELFHdr_e_shstrndx_Offset); // Set the NameIdx of each section in the string table and emit the bytes for // the string table. unsigned Index = 0; - DataBuffer &Buf = SHStrTab.SectionData; for (std::list::iterator I = SectionList.begin(), E = SectionList.end(); I != E; ++I) { // Set the index into the table. Note if we have lots of entries with // common suffixes, we could memoize them here if we cared. I->NameIdx = Index; - - // Add the name to the output buffer, including the null terminator. - Buf.insert(Buf.end(), I->Name.begin(), I->Name.end()); - - // Add a null terminator. - Buf.push_back(0); + SHStrTab.emitString(I->getName()); // Keep track of the number of bytes emitted to this section. - Index += I->Name.size()+1; + Index += I->getName().size()+1; } // Set the size of .shstrtab now that we know what it is. - assert(Index == Buf.size()); + assert(Index == SHStrTab.size()); SHStrTab.Size = Index; } @@ -391,9 +531,9 @@ void ELFWriter::EmitSectionTableStringTable() { /// SectionTable. void ELFWriter::OutputSectionsAndSectionTable() { // Pass #1: Compute the file offset for each section. - size_t FileOff = FileHeader.size(); // File header first. + size_t FileOff = ElfHdr.size(); // File header first. - // Emit all of the section data in order. + // Adjust alignment of all section if needed. for (std::list::iterator I = SectionList.begin(), E = SectionList.end(); I != E; ++I) { @@ -401,9 +541,14 @@ void ELFWriter::OutputSectionsAndSectionTable() { if (!I->SectionIdx) continue; + if (!I->size()) { + I->Offset = FileOff; + continue; + } + // Update Section size if (!I->Size) - I->Size = I->SectionData.size(); + I->Size = I->size(); // Align FileOff to whatever the alignment restrictions of the section are. if (I->Align) @@ -419,49 +564,40 @@ void ELFWriter::OutputSectionsAndSectionTable() { // Now that we know where all of the sections will be emitted, set the e_shnum // entry in the ELF header. - OutputBuffer FHOut(FileHeader, is64Bit, isLittleEndian); - FHOut.fixhalf(NumSections, ELFHdr_e_shnum_Offset); + ElfHdr.fixWord16(NumSections, ELFHdr_e_shnum_Offset); // Now that we know the offset in the file of the section table, update the // e_shoff address in the ELF header. - FHOut.fixaddr(FileOff, ELFHdr_e_shoff_Offset); + ElfHdr.fixWord(FileOff, ELFHdr_e_shoff_Offset); // Now that we know all of the data in the file header, emit it and all of the // sections! - O.write((char*)&FileHeader[0], FileHeader.size()); - FileOff = FileHeader.size(); - DataBuffer().swap(FileHeader); + O.write((char *)&ElfHdr.getData()[0], ElfHdr.size()); + FileOff = ElfHdr.size(); - DataBuffer Table; - OutputBuffer TableOut(Table, is64Bit, isLittleEndian); + // Section Header Table blob + BinaryObject SHdrTable(isLittleEndian, is64Bit); - // Emit all of the section data and build the section table itself. + // Emit all of sections to the file and build the section header table. while (!SectionList.empty()) { - const ELFSection &S = *SectionList.begin(); + ELFSection &S = *SectionList.begin(); + DOUT << "SectionIdx: " << S.SectionIdx << ", Name: " << S.getName() + << ", Size: " << S.Size << ", Offset: " << S.Offset + << ", SectionData Size: " << S.size() << "\n"; // Align FileOff to whatever the alignment restrictions of the section are. - if (S.Align) + if (S.Align) { for (size_t NewFileOff = (FileOff+S.Align-1) & ~(S.Align-1); - FileOff != NewFileOff; ++FileOff) + FileOff != NewFileOff; ++FileOff) O << (char)0xAB; - O.write((char*)&S.SectionData[0], S.Size); + } - DOUT << "SectionIdx: " << S.SectionIdx << ", Name: " << S.Name - << ", Size: " << S.Size << ", Offset: " << S.Offset << "\n"; - - FileOff += S.Size; - - TableOut.outword(S.NameIdx); // sh_name - Symbol table name idx - TableOut.outword(S.Type); // sh_type - Section contents & semantics - TableOut.outaddr(S.Flags); // sh_flags - Section flags. - TableOut.outaddr(S.Addr); // sh_addr - The mem addr this section is in. - TableOut.outaddr(S.Offset); // sh_offset - Offset from the file start. - TableOut.outaddr(S.Size); // sh_size - The section size. - TableOut.outword(S.Link); // sh_link - Section header table index link. - TableOut.outword(S.Info); // sh_info - Auxillary information. - TableOut.outaddr(S.Align); // sh_addralign - Alignment of section. - TableOut.outaddr(S.EntSize); // sh_entsize - Size of entries in the section + if (S.size()) { + O.write((char *)&S.getData()[0], S.Size); + FileOff += S.Size; + } + EmitSectionHeader(SHdrTable, S); SectionList.pop_front(); } @@ -471,5 +607,5 @@ void ELFWriter::OutputSectionsAndSectionTable() { O << (char)0xAB; // Emit the section table itself. - O.write((char*)&Table[0], Table.size()); + O.write((char *)&SHdrTable.getData()[0], SHdrTable.size()); } diff --git a/lib/CodeGen/ELFWriter.h b/lib/CodeGen/ELFWriter.h index 14a44f0240b..8a380f03401 100644 --- a/lib/CodeGen/ELFWriter.h +++ b/lib/CodeGen/ELFWriter.h @@ -16,15 +16,20 @@ #include "llvm/ADT/SetVector.h" #include "llvm/CodeGen/MachineFunctionPass.h" +#include "llvm/Support/OutputBuffer.h" +#include "llvm/Target/TargetAsmInfo.h" +#include "llvm/Target/TargetELFWriterInfo.h" #include "ELF.h" #include #include namespace llvm { + class BinaryObject; + class ConstantStruct; + class ELFCodeEmitter; class GlobalVariable; class Mangler; class MachineCodeEmitter; - class ELFCodeEmitter; class raw_ostream; /// ELFWriter - This class implements the common target-independent code for @@ -52,6 +57,9 @@ namespace llvm { /// Target machine description. TargetMachine &TM; + /// Target Elf Writer description. + const TargetELFWriterInfo *TEW; + /// Mang - The object used to perform name mangling for this module. Mangler *Mang; @@ -59,6 +67,10 @@ namespace llvm { /// code for functions to the .o file. ELFCodeEmitter *MCE; + /// TAI - Target Asm Info, provide information about section names for + /// globals and other target specific stuff. + const TargetAsmInfo *TAI; + //===------------------------------------------------------------------===// // Properties inferred automatically from the target machine. //===------------------------------------------------------------------===// @@ -77,13 +89,8 @@ namespace llvm { bool doFinalization(Module &M); private: - // The buffer we accumulate the file header into. Note that this should be - // changed into something much more efficient later (and the bitcode writer - // as well!). - DataBuffer FileHeader; - - /// ElfHdr - Hold information about the ELF Header - ELFHeader *ElfHdr; + // Blob containing the Elf header + BinaryObject ElfHdr; /// SectionList - This is the list of sections that we have emitted to the /// file. Once the file has been completely built, the section header table @@ -97,17 +104,18 @@ namespace llvm { /// getSection - Return the section with the specified name, creating a new /// section if one does not already exist. - ELFSection &getSection(const std::string &Name, - unsigned Type, unsigned Flags = 0) { + ELFSection &getSection(const std::string &Name, unsigned Type, + unsigned Flags = 0, unsigned Align = 0) { ELFSection *&SN = SectionLookup[Name]; if (SN) return *SN; - SectionList.push_back(Name); + SectionList.push_back(ELFSection(Name, isLittleEndian, is64Bit)); SN = &SectionList.back(); SN->SectionIdx = NumSections++; SN->Type = Type; SN->Flags = Flags; SN->Link = ELFSection::SHN_UNDEF; + SN->Align = Align; return *SN; } @@ -116,23 +124,36 @@ namespace llvm { ELFSection::SHF_EXECINSTR | ELFSection::SHF_ALLOC); } + ELFSection &getNonExecStackSection() { + return getSection(".note.GNU-stack", ELFSection::SHT_PROGBITS, 0, 1); + } + + ELFSection &getSymbolTableSection() { + return getSection(".symtab", ELFSection::SHT_SYMTAB, 0); + } + + ELFSection &getStringTableSection() { + return getSection(".strtab", ELFSection::SHT_STRTAB, 0, 1); + } + ELFSection &getDataSection() { return getSection(".data", ELFSection::SHT_PROGBITS, ELFSection::SHF_WRITE | ELFSection::SHF_ALLOC); } + ELFSection &getBSSSection() { return getSection(".bss", ELFSection::SHT_NOBITS, ELFSection::SHF_WRITE | ELFSection::SHF_ALLOC); } - /// SymbolTable - This is the list of symbols we have emitted to the file. + /// SymbolList - This is the list of symbols we have emitted to the file. /// This actually gets rearranged before emission to the file (to put the /// local symbols first in the list). - std::vector SymbolTable; + std::vector SymbolList; - /// PendingSyms - This is a list of externally defined symbols that we have - /// been asked to emit, but have not seen a reference to. When a reference - /// is seen, the symbol will move from this list to the SymbolTable. + /// PendingGlobals - List of externally defined symbols that we have been + /// asked to emit, but have not seen a reference to. When a reference + /// is seen, the symbol will move from this list to the SymbolList. SetVector PendingGlobals; // As we complete the ELF file, we need to update fields in the ELF header @@ -142,11 +163,17 @@ namespace llvm { unsigned ELFHdr_e_shoff_Offset; // e_shoff in ELF header. unsigned ELFHdr_e_shstrndx_Offset; // e_shstrndx in ELF header. unsigned ELFHdr_e_shnum_Offset; // e_shnum in ELF header. + private: void EmitGlobal(GlobalVariable *GV); - void EmitSymbolTable(); + void EmitGlobalConstant(const Constant *C, ELFSection &GblS); + void EmitGlobalConstantStruct(const ConstantStruct *CVS, + ELFSection &GblS); void EmitRelocations(); + void EmitSectionHeader(BinaryObject &SHdrTab, const ELFSection &SHdr); void EmitSectionTableStringTable(); + void EmitSymbol(BinaryObject &SymbolTable, ELFSym &Sym); + void EmitSymbolTable(); void OutputSectionsAndSectionTable(); }; } diff --git a/lib/CodeGen/LLVMTargetMachine.cpp b/lib/CodeGen/LLVMTargetMachine.cpp index b3c60e63932..a163cac75b0 100644 --- a/lib/CodeGen/LLVMTargetMachine.cpp +++ b/lib/CodeGen/LLVMTargetMachine.cpp @@ -240,7 +240,7 @@ bool LLVMTargetMachine::addCommonCodeGenPasses(PassManagerBase &PM, if (OptLevel != CodeGenOpt::None) { PM.add(createMachineLICMPass()); PM.add(createMachineSinkingPass()); - printAndVerify(PM, /* allowDoubleDefs= */ true); + printAndVerify(PM, /* allowDoubleDefs= */ false); } // Run pre-ra passes. diff --git a/lib/CodeGen/LazyLiveness.cpp b/lib/CodeGen/LazyLiveness.cpp new file mode 100644 index 00000000000..6fb35d235a7 --- /dev/null +++ b/lib/CodeGen/LazyLiveness.cpp @@ -0,0 +1,158 @@ +//===- LazyLiveness.cpp - Lazy, CFG-invariant liveness information --------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This pass implements a lazy liveness analysis as per "Fast Liveness Checking +// for SSA-form Programs," by Boissinot, et al. +// +//===----------------------------------------------------------------------===// + +#define DEBUG_TYPE "lazyliveness" +#include "llvm/CodeGen/LazyLiveness.h" +#include "llvm/CodeGen/MachineDominators.h" +#include "llvm/CodeGen/MachineRegisterInfo.h" +#include "llvm/CodeGen/Passes.h" +#include "llvm/ADT/DepthFirstIterator.h" +#include "llvm/ADT/PostOrderIterator.h" +using namespace llvm; + +char LazyLiveness::ID = 0; +static RegisterPass X("lazy-liveness", "Lazy Liveness Analysis"); + +void LazyLiveness::computeBackedgeChain(MachineFunction& mf, + MachineBasicBlock* MBB) { + SparseBitVector<128> tmp = rv[MBB]; + tmp.set(preorder[MBB]); + tmp &= backedge_source; + calculated.set(preorder[MBB]); + + for (SparseBitVector<128>::iterator I = tmp.begin(); I != tmp.end(); ++I) { + MachineBasicBlock* SrcMBB = rev_preorder[*I]; + + for (MachineBasicBlock::succ_iterator SI = SrcMBB->succ_begin(); + SI != SrcMBB->succ_end(); ++SI) { + MachineBasicBlock* TgtMBB = *SI; + + if (backedges.count(std::make_pair(SrcMBB, TgtMBB)) && + !rv[MBB].test(preorder[TgtMBB])) { + if (!calculated.test(preorder[TgtMBB])) + computeBackedgeChain(mf, TgtMBB); + + tv[MBB].set(preorder[TgtMBB]); + tv[MBB] |= tv[TgtMBB]; + } + } + + tv[MBB].reset(preorder[MBB]); + } +} + +bool LazyLiveness::runOnMachineFunction(MachineFunction &mf) { + rv.clear(); + tv.clear(); + backedges.clear(); + backedge_source.clear(); + backedge_target.clear(); + calculated.clear(); + preorder.clear(); + + MRI = &mf.getRegInfo(); + MachineDominatorTree& MDT = getAnalysis(); + + // Step 0: Compute preorder numbering for all MBBs. + unsigned num = 0; + for (df_iterator DI = df_begin(MDT.getRootNode()), + DE = df_end(MDT.getRootNode()); DI != DE; ++DI) { + preorder[(*DI)->getBlock()] = num++; + rev_preorder.push_back((*DI)->getBlock()); + } + + // Step 1: Compute the transitive closure of the CFG, ignoring backedges. + for (po_iterator POI = po_begin(&*mf.begin()), + POE = po_end(&*mf.begin()); POI != POE; ++POI) { + MachineBasicBlock* MBB = *POI; + SparseBitVector<128>& entry = rv[MBB]; + entry.set(preorder[MBB]); + + for (MachineBasicBlock::succ_iterator SI = MBB->succ_begin(), + SE = MBB->succ_end(); SI != SE; ++SI) { + DenseMap >::iterator SII = + rv.find(*SI); + + // Because we're iterating in postorder, any successor that does not yet + // have an rv entry must be on a backedge. + if (SII != rv.end()) { + entry |= SII->second; + } else { + backedges.insert(std::make_pair(MBB, *SI)); + backedge_source.set(preorder[MBB]); + backedge_target.set(preorder[*SI]); + } + } + } + + for (SparseBitVector<128>::iterator I = backedge_source.begin(); + I != backedge_source.end(); ++I) + computeBackedgeChain(mf, rev_preorder[*I]); + + for (po_iterator POI = po_begin(&*mf.begin()), + POE = po_end(&*mf.begin()); POI != POE; ++POI) + if (!backedge_target.test(preorder[*POI])) + for (MachineBasicBlock::succ_iterator SI = (*POI)->succ_begin(), + SE = (*POI)->succ_end(); SI != SE; ++SI) + if (!backedges.count(std::make_pair(*POI, *SI)) && tv.count(*SI)) { + SparseBitVector<128>& PBV = tv[*POI]; + PBV = tv[*SI]; + } + + for (po_iterator POI = po_begin(&*mf.begin()), + POE = po_end(&*mf.begin()); POI != POE; ++POI) + tv[*POI].set(preorder[*POI]); + + return false; +} + +bool LazyLiveness::vregLiveIntoMBB(unsigned vreg, MachineBasicBlock* MBB) { + MachineDominatorTree& MDT = getAnalysis(); + + MachineBasicBlock* DefMBB = MRI->def_begin(vreg)->getParent(); + unsigned def = preorder[DefMBB]; + unsigned max_dom = 0; + for (df_iterator DI = df_begin(MDT[DefMBB]), + DE = df_end(MDT[DefMBB]); DI != DE; ++DI) { + if (preorder[DI->getBlock()] > max_dom) { + max_dom = preorder[(*DI)->getBlock()]; + } + } + + if (preorder[MBB] <= def || max_dom < preorder[MBB]) + return false; + + SparseBitVector<128>::iterator I = tv[MBB].begin(); + while (I != tv[MBB].end() && *I <= def) ++I; + while (I != tv[MBB].end() && *I < max_dom) { + for (MachineRegisterInfo::use_iterator UI = MRI->use_begin(vreg), + UE = MachineRegisterInfo::use_end(); UI != UE; ++UI) { + MachineBasicBlock* UseMBB = UI->getParent(); + if (rv[rev_preorder[*I]].test(preorder[UseMBB])) + return true; + + unsigned t_dom = 0; + for (df_iterator DI = + df_begin(MDT[rev_preorder[*I]]), DE = df_end(MDT[rev_preorder[*I]]); + DI != DE; ++DI) + if (preorder[DI->getBlock()] > t_dom) { + max_dom = preorder[(*DI)->getBlock()]; + } + I = tv[MBB].begin(); + while (I != tv[MBB].end() && *I < t_dom) ++I; + } + } + + return false; +} diff --git a/lib/CodeGen/RegisterScavenging.cpp b/lib/CodeGen/RegisterScavenging.cpp index 944468ea11b..3feb92fc46d 100644 --- a/lib/CodeGen/RegisterScavenging.cpp +++ b/lib/CodeGen/RegisterScavenging.cpp @@ -214,26 +214,33 @@ void RegScavenger::forward() { } // Process uses first. - BitVector UseRegs(NumPhysRegs); + BitVector KillRegs(NumPhysRegs); for (unsigned i = 0, e = UseMOs.size(); i != e; ++i) { const MachineOperand MO = *UseMOs[i].first; unsigned Reg = MO.getReg(); assert(isUsed(Reg) && "Using an undefined register!"); - if (MO.isKill() && !isReserved(Reg)) { - UseRegs.set(Reg); + // Kill of implicit_def defined registers are ignored. e.g. + // entry: 0x2029ab8, LLVM BB @0x1b06080, ID#0: + // Live Ins: %R0 + // %R0 = IMPLICIT_DEF + // %R0 = IMPLICIT_DEF + // STR %R0, %R0, %reg0, 0, 14, %reg0, Mem:ST(4,4) [0x1b06510 + 0] + // %R1 = LDR %R0, %reg0, 24, 14, %reg0, Mem:LD(4,4) [0x1b065bc + 0] + if (MO.isKill() && !isReserved(Reg) && !isImplicitlyDefined(Reg)) { + KillRegs.set(Reg); // Mark sub-registers as used. for (const unsigned *SubRegs = TRI->getSubRegisters(Reg); unsigned SubReg = *SubRegs; ++SubRegs) - UseRegs.set(SubReg); + KillRegs.set(SubReg); } } // Change states of all registers after all the uses are processed to guard // against multiple uses. - setUnused(UseRegs); + setUnused(KillRegs); // Process early clobber defs then process defs. We can have a early clobber // that is dead, it should not conflict with a def that happens one "slot" diff --git a/lib/ExecutionEngine/JIT/JIT.cpp b/lib/ExecutionEngine/JIT/JIT.cpp index f8ae8844616..43995cb0ecd 100644 --- a/lib/ExecutionEngine/JIT/JIT.cpp +++ b/lib/ExecutionEngine/JIT/JIT.cpp @@ -563,6 +563,11 @@ void *JIT::getPointerToFunction(Function *F) { return Addr; // Check if function already code gen'd MutexGuard locked(lock); + + // Now that this thread owns the lock, check if another thread has already + // code gen'd the function. + if (void *Addr = getPointerToGlobalIfAvailable(F)) + return Addr; // Make sure we read in the function if it exists in this Module. if (F->hasNotBeenReadFromBitcode()) { diff --git a/lib/Target/ARM/ARM.h b/lib/Target/ARM/ARM.h index ac7de911cee..7edd1180f27 100644 --- a/lib/Target/ARM/ARM.h +++ b/lib/Target/ARM/ARM.h @@ -98,12 +98,12 @@ FunctionPass *createARMCodePrinterPass(raw_ostream &O, FunctionPass *createARMCodeEmitterPass(ARMTargetMachine &TM, MachineCodeEmitter &MCE); -FunctionPass *createARMCodeEmitterPass( ARMTargetMachine &TM, - MachineCodeEmitter &MCE); -FunctionPass *createARMJITCodeEmitterPass( ARMTargetMachine &TM, - JITCodeEmitter &JCE); +FunctionPass *createARMCodeEmitterPass(ARMTargetMachine &TM, + MachineCodeEmitter &MCE); +FunctionPass *createARMJITCodeEmitterPass(ARMTargetMachine &TM, + JITCodeEmitter &JCE); -FunctionPass *createARMLoadStoreOptimizationPass(); +FunctionPass *createARMLoadStoreOptimizationPass(bool PreAlloc = false); FunctionPass *createARMConstantIslandPass(); } // end namespace llvm; diff --git a/lib/Target/ARM/ARM.td b/lib/Target/ARM/ARM.td index 4ac6857d22d..594811d6357 100644 --- a/lib/Target/ARM/ARM.td +++ b/lib/Target/ARM/ARM.td @@ -28,6 +28,8 @@ def ArchV5TE : SubtargetFeature<"v5te", "ARMArchVersion", "V5TE", "ARM v5TE, v5TEj, v5TExp">; def ArchV6 : SubtargetFeature<"v6", "ARMArchVersion", "V6", "ARM v6">; +def ArchV6T2 : SubtargetFeature<"v6t2", "ARMArchVersion", "V6T2", + "ARM v6t2">; def ArchV7A : SubtargetFeature<"v7a", "ARMArchVersion", "V7A", "ARM v7A">; def FeatureVFP2 : SubtargetFeature<"vfp2", "ARMFPUType", "VFPv2", @@ -92,9 +94,11 @@ def : Proc<"arm1176jzf-s", [ArchV6, FeatureVFP2]>; def : Proc<"mpcorenovfp", [ArchV6]>; def : Proc<"mpcore", [ArchV6, FeatureVFP2]>; -def : Proc<"arm1156t2-s", [ArchV6, FeatureThumb2]>; -def : Proc<"arm1156t2f-s", [ArchV6, FeatureThumb2, FeatureVFP2]>; +// V6T2 Processors. +def : Proc<"arm1156t2-s", [ArchV6T2, FeatureThumb2]>; +def : Proc<"arm1156t2f-s", [ArchV6T2, FeatureThumb2, FeatureVFP2]>; +// V7 Processors. def : Proc<"cortex-a8", [ArchV7A, FeatureThumb2, FeatureNEON]>; def : Proc<"cortex-a9", [ArchV7A, FeatureThumb2, FeatureNEON]>; diff --git a/lib/Target/ARM/ARMCallingConv.td b/lib/Target/ARM/ARMCallingConv.td index 6cd786eed4d..f126760cd5e 100644 --- a/lib/Target/ARM/ARMCallingConv.td +++ b/lib/Target/ARM/ARMCallingConv.td @@ -17,6 +17,11 @@ class CCIfSubtarget: class CCIfAlign: CCIf; +/// CCIfFloatABI - Match of the float ABI and the arg. ABIType may be "Hard" or +/// "Soft". +class CCIfFloatABI: + CCIf; + //===----------------------------------------------------------------------===// // ARM APCS Calling Convention //===----------------------------------------------------------------------===// @@ -43,9 +48,10 @@ def RetCC_ARM_APCS : CallingConv<[ ]>; //===----------------------------------------------------------------------===// -// ARM AAPCS (EABI) Calling Convention +// ARM AAPCS (EABI) Calling Convention, common parts //===----------------------------------------------------------------------===// -def CC_ARM_AAPCS : CallingConv<[ + +def CC_ARM_AAPCS_Common : CallingConv<[ CCIfType<[i8, i16], CCPromoteToType>, @@ -53,35 +59,71 @@ def CC_ARM_AAPCS : CallingConv<[ // i64 is 8-aligned i32 here, so we may need to eat R1 as a pad register // (and the same is true for f64 if VFP is not enabled) CCIfType<[i32], CCIfAlign<"8", CCAssignToRegWithShadow<[R0, R2], [R0, R1]>>>, - CCIfType<[f64], CCCustom<"CC_ARM_AAPCS_Custom_f64">>, - - CCIfType<[f32], CCBitConvertToType>, CCIfType<[i32], CCIf<"State.getNextStackOffset() == 0 &&" "ArgFlags.getOrigAlign() != 8", CCAssignToReg<[R0, R1, R2, R3]>>>, - CCIfType<[i32], CCAssignToStack<4, 4>>, + CCIfType<[i32, f32], CCAssignToStack<4, 4>>, CCIfType<[f64], CCAssignToStack<8, 8>> ]>; -def RetCC_ARM_AAPCS : CallingConv<[ - CCIfType<[f32], CCBitConvertToType>, - CCIfType<[f64], CCCustom<"RetCC_ARM_AAPCS_Custom_f64">>, - +def RetCC_ARM_AAPCS_Common : CallingConv<[ CCIfType<[i32], CCAssignToReg<[R0, R1, R2, R3]>>, CCIfType<[i64], CCAssignToRegWithShadow<[R0, R2], [R1, R3]>> ]>; +//===----------------------------------------------------------------------===// +// ARM AAPCS (EABI) Calling Convention +//===----------------------------------------------------------------------===// + +def CC_ARM_AAPCS : CallingConv<[ + CCIfType<[f64], CCCustom<"CC_ARM_AAPCS_Custom_f64">>, + CCIfType<[f32], CCBitConvertToType>, + CCDelegateTo +]>; + +def RetCC_ARM_AAPCS : CallingConv<[ + CCIfType<[f64], CCCustom<"RetCC_ARM_AAPCS_Custom_f64">>, + CCIfType<[f32], CCBitConvertToType>, + CCDelegateTo +]>; + +//===----------------------------------------------------------------------===// +// ARM AAPCS-VFP (EABI) Calling Convention +//===----------------------------------------------------------------------===// + +def CC_ARM_AAPCS_VFP : CallingConv<[ + CCIfType<[f64], CCAssignToReg<[D0, D1, D2, D3, D4, D5, D6, D7]>>, + CCIfType<[f32], CCAssignToReg<[S0, S1, S2, S3, S4, S5, S6, S7, S8, + S9, S10, S11, S12, S13, S14, S15]>>, + CCDelegateTo +]>; + +def RetCC_ARM_AAPCS_VFP : CallingConv<[ + CCIfType<[f64], CCAssignToReg<[D0, D1, D2, D3, D4, D5, D6, D7]>>, + CCIfType<[f32], CCAssignToReg<[S0, S1, S2, S3, S4, S5, S6, S7, S8, + S9, S10, S11, S12, S13, S14, S15]>>, + CCDelegateTo +]>; + //===----------------------------------------------------------------------===// // ARM Calling Convention Dispatch //===----------------------------------------------------------------------===// def CC_ARM : CallingConv<[ + CCIfSubtarget<"isAAPCS_ABI()", + CCIfSubtarget<"hasVFP2()", + CCIfFloatABI<"Hard", + CCDelegateTo>>>, CCIfSubtarget<"isAAPCS_ABI()", CCDelegateTo>, CCDelegateTo ]>; def RetCC_ARM : CallingConv<[ + CCIfSubtarget<"isAAPCS_ABI()", + CCIfSubtarget<"hasVFP2()", + CCIfFloatABI<"Hard", + CCDelegateTo>>>, CCIfSubtarget<"isAAPCS_ABI()", CCDelegateTo>, CCDelegateTo ]>; diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp index c0fd9dcd107..ec8bd1f8b31 100644 --- a/lib/Target/ARM/ARMISelLowering.cpp +++ b/lib/Target/ARM/ARMISelLowering.cpp @@ -1101,7 +1101,12 @@ ARMTargetLowering::LowerFORMAL_ARGUMENTS(SDValue Op, SelectionDAG &DAG) { else RC = ARM::GPRRegisterClass; - if (RegVT == MVT::f64) { + if (FloatABIType == FloatABI::Hard) { + if (RegVT == MVT::f32) + RC = ARM::SPRRegisterClass; + else if (RegVT == MVT::f64) + RC = ARM::DPRRegisterClass; + } else if (RegVT == MVT::f64) { // f64 is passed in pairs of GPRs and must be combined. RegVT = MVT::i32; } else if (!((RegVT == MVT::i32) || (RegVT == MVT::f32))) diff --git a/lib/Target/ARM/ARMInstrInfo.td b/lib/Target/ARM/ARMInstrInfo.td index 680e77234db..cc9f1a5759d 100644 --- a/lib/Target/ARM/ARMInstrInfo.td +++ b/lib/Target/ARM/ARMInstrInfo.td @@ -451,7 +451,7 @@ multiclass AsXI1_bin_c_irs opcod, string opc, PatFrag opnode> { /// the function. The first operand is the ID# for this instruction, the second /// is the index into the MachineConstantPool that this is, the third is the /// size in bytes of this constant pool entry. -let isNotDuplicable = 1 in +let neverHasSideEffects = 1, isNotDuplicable = 1 in def CONSTPOOL_ENTRY : PseudoInst<(outs), (ins cpinst_operand:$instid, cpinst_operand:$cpidx, i32imm:$size), @@ -771,6 +771,7 @@ def STM : AXI4st<(outs), // Move Instructions. // +let neverHasSideEffects = 1 in def MOVr : AsI1<0b1101, (outs GPR:$dst), (ins GPR:$src), DPFrm, "mov", " $dst, $src", []>, UnaryDP; def MOVs : AsI1<0b1101, (outs GPR:$dst), (ins so_reg:$src), DPSoRegFrm, @@ -946,6 +947,7 @@ def MLA : AsMul1I<0b0000001, (outs GPR:$dst), (ins GPR:$a, GPR:$b, GPR:$c), [(set GPR:$dst, (add (mul GPR:$a, GPR:$b), GPR:$c))]>; // Extra precision multiplies with low / high results +let neverHasSideEffects = 1 in { def SMULL : AsMul1I<0b0000110, (outs GPR:$ldst, GPR:$hdst), (ins GPR:$a, GPR:$b), "smull", " $ldst, $hdst, $a, $b", []>; @@ -967,6 +969,7 @@ def UMAAL : AMul1I <0b0000010, (outs GPR:$ldst, GPR:$hdst), (ins GPR:$a, GPR:$b), "umaal", " $ldst, $hdst, $a, $b", []>, Requires<[IsARM, HasV6]>; +} // neverHasSideEffects // Most significant word multiply def SMMUL : AMul2I <0b0111010, (outs GPR:$dst), (ins GPR:$a, GPR:$b), diff --git a/lib/Target/ARM/ARMInstrThumb.td b/lib/Target/ARM/ARMInstrThumb.td index ffb83a8b4d3..54232f6dfe6 100644 --- a/lib/Target/ARM/ARMInstrThumb.td +++ b/lib/Target/ARM/ARMInstrThumb.td @@ -298,6 +298,7 @@ def tADDrr : TI<(outs tGPR:$dst), (ins tGPR:$lhs, tGPR:$rhs), "add $dst, $lhs, $rhs", [(set tGPR:$dst, (add tGPR:$lhs, tGPR:$rhs))]>; +let neverHasSideEffects = 1 in def tADDhirr : TIt<(outs tGPR:$dst), (ins GPR:$lhs, GPR:$rhs), "add $dst, $rhs @ addhirr", []>; @@ -387,6 +388,7 @@ def tMOVi8 : TI<(outs tGPR:$dst), (ins i32imm:$src), // Note: MOV(2) of two low regs updates the flags, so we emit this as 'cpy', // which is MOV(3). This also supports high registers. +let neverHasSideEffects = 1 in { def tMOVr : TI<(outs tGPR:$dst), (ins tGPR:$src), "cpy $dst, $src", []>; def tMOVhir2lor : TI<(outs tGPR:$dst), (ins GPR:$src), @@ -395,6 +397,7 @@ def tMOVlor2hir : TI<(outs GPR:$dst), (ins tGPR:$src), "cpy $dst, $src\t@ lor2hir", []>; def tMOVhir2hir : TI<(outs GPR:$dst), (ins GPR:$src), "cpy $dst, $src\t@ hir2hir", []>; +} // neverHasSideEffects def tMUL : TIt<(outs tGPR:$dst), (ins tGPR:$lhs, tGPR:$rhs), "mul $dst, $rhs", diff --git a/lib/Target/ARM/ARMInstrVFP.td b/lib/Target/ARM/ARMInstrVFP.td index 0247dafd07d..9104c77115f 100644 --- a/lib/Target/ARM/ARMInstrVFP.td +++ b/lib/Target/ARM/ARMInstrVFP.td @@ -192,11 +192,13 @@ def FCVTSD : AI<(outs SPR:$dst), (ins DPR:$a), VFPUnaryFrm, let Inst{7-4} = 0b1100; } +let neverHasSideEffects = 1 in { def FCPYD : ADuI<0b11101011, 0b0000, 0b0100, (outs DPR:$dst), (ins DPR:$a), "fcpyd", " $dst, $a", []>; def FCPYS : ASuI<0b11101011, 0b0000, 0b0100, (outs SPR:$dst), (ins SPR:$a), "fcpys", " $dst, $a", []>; +} // neverHasSideEffects def FNEGD : ADuI<0b11101011, 0b0001, 0b0100, (outs DPR:$dst), (ins DPR:$a), "fnegd", " $dst, $a", diff --git a/lib/Target/ARM/ARMLoadStoreOptimizer.cpp b/lib/Target/ARM/ARMLoadStoreOptimizer.cpp index 963ff0def64..684ecb4ce79 100644 --- a/lib/Target/ARM/ARMLoadStoreOptimizer.cpp +++ b/lib/Target/ARM/ARMLoadStoreOptimizer.cpp @@ -17,24 +17,31 @@ #include "ARMAddressingModes.h" #include "ARMMachineFunctionInfo.h" #include "ARMRegisterInfo.h" -#include "llvm/ADT/STLExtras.h" -#include "llvm/ADT/SmallVector.h" -#include "llvm/ADT/Statistic.h" #include "llvm/CodeGen/MachineBasicBlock.h" #include "llvm/CodeGen/MachineFunctionPass.h" #include "llvm/CodeGen/MachineInstr.h" #include "llvm/CodeGen/MachineInstrBuilder.h" +#include "llvm/CodeGen/MachineRegisterInfo.h" #include "llvm/CodeGen/RegisterScavenging.h" -#include "llvm/Support/Compiler.h" #include "llvm/Target/TargetRegisterInfo.h" #include "llvm/Target/TargetInstrInfo.h" #include "llvm/Target/TargetMachine.h" +#include "llvm/Support/Compiler.h" +#include "llvm/ADT/DenseMap.h" +#include "llvm/ADT/STLExtras.h" +#include "llvm/ADT/SmallPtrSet.h" +#include "llvm/ADT/SmallVector.h" +#include "llvm/ADT/Statistic.h" using namespace llvm; STATISTIC(NumLDMGened , "Number of ldm instructions generated"); STATISTIC(NumSTMGened , "Number of stm instructions generated"); STATISTIC(NumFLDMGened, "Number of fldm instructions generated"); STATISTIC(NumFSTMGened, "Number of fstm instructions generated"); +STATISTIC(NumLdStMoved, "Number of load / store instructions moved"); + +/// ARMAllocLoadStoreOpt - Post- register allocation pass the combine +/// load / store instructions to form ldm / stm instructions. namespace { struct VISIBILITY_HIDDEN ARMLoadStoreOpt : public MachineFunctionPass { @@ -81,12 +88,6 @@ namespace { char ARMLoadStoreOpt::ID = 0; } -/// createARMLoadStoreOptimizationPass - returns an instance of the load / store -/// optimization pass. -FunctionPass *llvm::createARMLoadStoreOptimizationPass() { - return new ARMLoadStoreOpt(); -} - static int getLoadStoreMultipleOpcode(int Opcode) { switch (Opcode) { case ARM::LDR: @@ -582,6 +583,23 @@ void ARMLoadStoreOpt::AdvanceRS(MachineBasicBlock &MBB, MemOpQueue &MemOps) { RS->forward(prior(Loc)); } +static int getMemoryOpOffset(const MachineInstr *MI) { + int Opcode = MI->getOpcode(); + bool isAM2 = Opcode == ARM::LDR || Opcode == ARM::STR; + unsigned NumOperands = MI->getDesc().getNumOperands(); + unsigned OffField = MI->getOperand(NumOperands-3).getImm(); + int Offset = isAM2 + ? ARM_AM::getAM2Offset(OffField) : ARM_AM::getAM5Offset(OffField) * 4; + if (isAM2) { + if (ARM_AM::getAM2Op(OffField) == ARM_AM::sub) + Offset = -Offset; + } else { + if (ARM_AM::getAM5Op(OffField) == ARM_AM::sub) + Offset = -Offset; + } + return Offset; +} + /// LoadStoreMultipleOpti - An optimization pass to turn multiple LDR / STR /// ops of the same base and incrementing offset into LDM / STM ops. bool ARMLoadStoreOpt::LoadStoreMultipleOpti(MachineBasicBlock &MBB) { @@ -606,22 +624,11 @@ bool ARMLoadStoreOpt::LoadStoreMultipleOpti(MachineBasicBlock &MBB) { bool isMemOp = isMemoryOp(MBBI); if (isMemOp) { int Opcode = MBBI->getOpcode(); - bool isAM2 = Opcode == ARM::LDR || Opcode == ARM::STR; unsigned Size = getLSMultipleTransferSize(MBBI); unsigned Base = MBBI->getOperand(1).getReg(); unsigned PredReg = 0; ARMCC::CondCodes Pred = getInstrPredicate(MBBI, PredReg); - unsigned NumOperands = MBBI->getDesc().getNumOperands(); - unsigned OffField = MBBI->getOperand(NumOperands-3).getImm(); - int Offset = isAM2 - ? ARM_AM::getAM2Offset(OffField) : ARM_AM::getAM5Offset(OffField) * 4; - if (isAM2) { - if (ARM_AM::getAM2Op(OffField) == ARM_AM::sub) - Offset = -Offset; - } else { - if (ARM_AM::getAM5Op(OffField) == ARM_AM::sub) - Offset = -Offset; - } + int Offset = getMemoryOpOffset(MBBI); // Watch out for: // r4 := ldr [r5] // r5 := ldr [r5, #4] @@ -744,6 +751,17 @@ bool ARMLoadStoreOpt::LoadStoreMultipleOpti(MachineBasicBlock &MBB) { return NumMerges > 0; } +namespace { + struct OffsetCompare { + bool operator()(const MachineInstr *LHS, const MachineInstr *RHS) const { + int LOffset = getMemoryOpOffset(LHS); + int ROffset = getMemoryOpOffset(RHS); + assert(LHS == RHS || LOffset != ROffset); + return LOffset > ROffset; + } + }; +} + /// MergeReturnIntoLDM - If this is a exit BB, try merging the return op /// (bx lr) into the preceeding stack restore so it directly restore the value /// of LR into pc. @@ -788,3 +806,277 @@ bool ARMLoadStoreOpt::runOnMachineFunction(MachineFunction &Fn) { delete RS; return Modified; } + + +/// ARMPreAllocLoadStoreOpt - Pre- register allocation pass that move +/// load / stores from consecutive locations close to make it more +/// likely they will be combined later. + +namespace { + struct VISIBILITY_HIDDEN ARMPreAllocLoadStoreOpt : public MachineFunctionPass{ + static char ID; + ARMPreAllocLoadStoreOpt() : MachineFunctionPass(&ID) {} + + const TargetInstrInfo *TII; + const TargetRegisterInfo *TRI; + MachineRegisterInfo *MRI; + + virtual bool runOnMachineFunction(MachineFunction &Fn); + + virtual const char *getPassName() const { + return "ARM pre- register allocation load / store optimization pass"; + } + + private: + bool RescheduleOps(MachineBasicBlock *MBB, + SmallVector &Ops, + unsigned Base, bool isLd, + DenseMap &MI2LocMap); + bool RescheduleLoadStoreInstrs(MachineBasicBlock *MBB); + }; + char ARMPreAllocLoadStoreOpt::ID = 0; +} + +bool ARMPreAllocLoadStoreOpt::runOnMachineFunction(MachineFunction &Fn) { + TII = Fn.getTarget().getInstrInfo(); + TRI = Fn.getTarget().getRegisterInfo(); + MRI = &Fn.getRegInfo(); + + bool Modified = false; + for (MachineFunction::iterator MFI = Fn.begin(), E = Fn.end(); MFI != E; + ++MFI) + Modified |= RescheduleLoadStoreInstrs(MFI); + + return Modified; +} + +static bool IsSafeToMove(bool isLd, unsigned Base, + MachineBasicBlock::iterator I, + MachineBasicBlock::iterator E, + SmallPtrSet MoveOps, + const TargetRegisterInfo *TRI) { + // Are there stores / loads / calls between them? + // FIXME: This is overly conservative. We should make use of alias information + // some day. + while (++I != E) { + const TargetInstrDesc &TID = I->getDesc(); + if (TID.isCall() || TID.isTerminator() || TID.hasUnmodeledSideEffects()) + return false; + if (isLd && TID.mayStore()) + return false; + if (!isLd) { + if (TID.mayLoad()) + return false; + // It's not safe to move the first 'str' down. + // str r1, [r0] + // strh r5, [r0] + // str r4, [r0, #+4] + if (TID.mayStore() && !MoveOps.count(&*I)) + return false; + } + for (unsigned j = 0, NumOps = I->getNumOperands(); j != NumOps; ++j) { + MachineOperand &MO = I->getOperand(j); + if (MO.isReg() && MO.isDef() && TRI->regsOverlap(MO.getReg(), Base)) + return false; + } + } + return true; +} + +bool ARMPreAllocLoadStoreOpt::RescheduleOps(MachineBasicBlock *MBB, + SmallVector &Ops, + unsigned Base, bool isLd, + DenseMap &MI2LocMap) { + bool RetVal = false; + + // Sort by offset (in reverse order). + std::sort(Ops.begin(), Ops.end(), OffsetCompare()); + + // The loads / stores of the same base are in order. Scan them from first to + // last and check for the followins: + // 1. Any def of base. + // 2. Any gaps. + while (Ops.size() > 1) { + unsigned FirstLoc = ~0U; + unsigned LastLoc = 0; + MachineInstr *FirstOp = 0; + MachineInstr *LastOp = 0; + int LastOffset = 0; + unsigned LastBytes = 0; + unsigned NumMove = 0; + for (int i = Ops.size() - 1; i >= 0; --i) { + MachineInstr *Op = Ops[i]; + unsigned Loc = MI2LocMap[Op]; + if (Loc <= FirstLoc) { + FirstLoc = Loc; + FirstOp = Op; + } + if (Loc >= LastLoc) { + LastLoc = Loc; + LastOp = Op; + } + + int Offset = getMemoryOpOffset(Op); + unsigned Bytes = getLSMultipleTransferSize(Op); + if (LastBytes) { + if (Bytes != LastBytes || Offset != (LastOffset + (int)Bytes)) + break; + } + LastOffset = Offset; + LastBytes = Bytes; + if (++NumMove == 4) + break; + } + + if (NumMove <= 1) + Ops.pop_back(); + else { + SmallPtrSet MoveOps; + for (int i = NumMove-1; i >= 0; --i) + MoveOps.insert(Ops[i]); + + // Be conservative, if the instructions are too far apart, don't + // move them. We want to limit the increase of register pressure. + bool DoMove = (LastLoc - FirstLoc) < NumMove*4; + if (DoMove) + DoMove = IsSafeToMove(isLd, Base, FirstOp, LastOp, MoveOps, TRI); + if (!DoMove) { + for (unsigned i = 0; i != NumMove; ++i) + Ops.pop_back(); + } else { + // This is the new location for the loads / stores. + MachineBasicBlock::iterator InsertPos = isLd ? FirstOp : LastOp; + while (InsertPos != MBB->end() && MoveOps.count(InsertPos)) + ++InsertPos; + for (unsigned i = 0; i != NumMove; ++i) { + MachineInstr *Op = Ops.back(); + Ops.pop_back(); + MBB->splice(InsertPos, MBB, Op); + } + + NumLdStMoved += NumMove; + RetVal = true; + } + } + } + + return RetVal; +} + +bool +ARMPreAllocLoadStoreOpt::RescheduleLoadStoreInstrs(MachineBasicBlock *MBB) { + bool RetVal = false; + + DenseMap MI2LocMap; + DenseMap > Base2LdsMap; + DenseMap > Base2StsMap; + SmallVector LdBases; + SmallVector StBases; + + unsigned Loc = 0; + MachineBasicBlock::iterator MBBI = MBB->begin(); + MachineBasicBlock::iterator E = MBB->end(); + while (MBBI != E) { + for (; MBBI != E; ++MBBI) { + MachineInstr *MI = MBBI; + const TargetInstrDesc &TID = MI->getDesc(); + if (TID.isCall() || TID.isTerminator()) { + // Stop at barriers. + ++MBBI; + break; + } + + MI2LocMap[MI] = Loc++; + if (!isMemoryOp(MI)) + continue; + unsigned PredReg = 0; + if (getInstrPredicate(MI, PredReg) != ARMCC::AL) + continue; + + int Opcode = MI->getOpcode(); + bool isLd = Opcode == ARM::LDR || + Opcode == ARM::FLDS || Opcode == ARM::FLDD; + unsigned Base = MI->getOperand(1).getReg(); + int Offset = getMemoryOpOffset(MI); + + bool StopHere = false; + if (isLd) { + DenseMap >::iterator BI = + Base2LdsMap.find(Base); + if (BI != Base2LdsMap.end()) { + for (unsigned i = 0, e = BI->second.size(); i != e; ++i) { + if (Offset == getMemoryOpOffset(BI->second[i])) { + StopHere = true; + break; + } + } + if (!StopHere) + BI->second.push_back(MI); + } else { + SmallVector MIs; + MIs.push_back(MI); + Base2LdsMap[Base] = MIs; + LdBases.push_back(Base); + } + } else { + DenseMap >::iterator BI = + Base2StsMap.find(Base); + if (BI != Base2StsMap.end()) { + for (unsigned i = 0, e = BI->second.size(); i != e; ++i) { + if (Offset == getMemoryOpOffset(BI->second[i])) { + StopHere = true; + break; + } + } + if (!StopHere) + BI->second.push_back(MI); + } else { + SmallVector MIs; + MIs.push_back(MI); + Base2StsMap[Base] = MIs; + StBases.push_back(Base); + } + } + + if (StopHere) { + // Found a duplicate (a base+offset combination that's seen earlier). Backtrack. + --Loc; + break; + } + } + + // Re-schedule loads. + for (unsigned i = 0, e = LdBases.size(); i != e; ++i) { + unsigned Base = LdBases[i]; + SmallVector &Lds = Base2LdsMap[Base]; + if (Lds.size() > 1) + RetVal |= RescheduleOps(MBB, Lds, Base, true, MI2LocMap); + } + + // Re-schedule stores. + for (unsigned i = 0, e = StBases.size(); i != e; ++i) { + unsigned Base = StBases[i]; + SmallVector &Sts = Base2StsMap[Base]; + if (Sts.size() > 1) + RetVal |= RescheduleOps(MBB, Sts, Base, false, MI2LocMap); + } + + if (MBBI != E) { + Base2LdsMap.clear(); + Base2StsMap.clear(); + LdBases.clear(); + StBases.clear(); + } + } + + return RetVal; +} + + +/// createARMLoadStoreOptimizationPass - returns an instance of the load / store +/// optimization pass. +FunctionPass *llvm::createARMLoadStoreOptimizationPass(bool PreAlloc) { + if (PreAlloc) + return new ARMPreAllocLoadStoreOpt(); + return new ARMLoadStoreOpt(); +} diff --git a/lib/Target/ARM/ARMRegisterInfo.td b/lib/Target/ARM/ARMRegisterInfo.td index b95d1f93aba..ebe7d582a66 100644 --- a/lib/Target/ARM/ARMRegisterInfo.td +++ b/lib/Target/ARM/ARMRegisterInfo.td @@ -219,3 +219,18 @@ def DPR : RegisterClass<"ARM", [f64], 64, [D0, D1, D2, D3, D4, D5, D6, D7, D8, // Condition code registers. def CCR : RegisterClass<"ARM", [i32], 32, [CPSR]>; + +//===----------------------------------------------------------------------===// +// Subregister Set Definitions... now that we have all of the pieces, define the +// sub registers for each register. +// + +def : SubRegSet<1, [D0, D1, D2, D3, D4, D5, D6, D7, + D8, D9, D10, D11, D12, D13, D14, D15], + [S0, S2, S4, S6, S8, S10, S12, S14, + S16, S18, S20, S22, S24, S26, S28, S30]>; + +def : SubRegSet<2, [D0, D1, D2, D3, D4, D5, D6, D7, + D8, D9, D10, D11, D12, D13, D14, D15], + [S1, S3, S5, S7, S9, S11, S13, S15, + S17, S19, S21, S23, S25, S27, S29, S31]>; diff --git a/lib/Target/ARM/ARMSubtarget.cpp b/lib/Target/ARM/ARMSubtarget.cpp index ef78cd52d85..a978380b627 100644 --- a/lib/Target/ARM/ARMSubtarget.cpp +++ b/lib/Target/ARM/ARMSubtarget.cpp @@ -14,6 +14,8 @@ #include "ARMSubtarget.h" #include "ARMGenSubtarget.inc" #include "llvm/Module.h" +#include "llvm/Target/TargetMachine.h" +#include "llvm/Target/TargetOptions.h" using namespace llvm; ARMSubtarget::ARMSubtarget(const Module &M, const std::string &FS, @@ -28,6 +30,10 @@ ARMSubtarget::ARMSubtarget(const Module &M, const std::string &FS, , CPUString("generic") , TargetType(isELF) // Default to ELF unless otherwise specified. , TargetABI(ARM_ABI_APCS) { + // default to soft float ABI + if (FloatABIType == FloatABI::Default) + FloatABIType = FloatABI::Soft; + // Determine default and user specified characteristics // Parse features string. diff --git a/lib/Target/ARM/ARMSubtarget.h b/lib/Target/ARM/ARMSubtarget.h index 8b469cff77d..07040558d87 100644 --- a/lib/Target/ARM/ARMSubtarget.h +++ b/lib/Target/ARM/ARMSubtarget.h @@ -23,7 +23,7 @@ class Module; class ARMSubtarget : public TargetSubtarget { protected: enum ARMArchEnum { - V4T, V5T, V5TE, V6, V7A + V4T, V5T, V5TE, V6, V6T2, V7A }; enum ARMFPEnum { @@ -92,6 +92,7 @@ protected: bool hasV5TOps() const { return ARMArchVersion >= V5T; } bool hasV5TEOps() const { return ARMArchVersion >= V5TE; } bool hasV6Ops() const { return ARMArchVersion >= V6; } + bool hasV6T2Ops() const { return ARMArchVersion >= V6T2; } bool hasV7Ops() const { return ARMArchVersion >= V7A; } bool hasVFP2() const { return ARMFPUType >= VFPv2; } @@ -105,6 +106,7 @@ protected: bool isAAPCS_ABI() const { return TargetABI == ARM_ABI_AAPCS; } bool isThumb() const { return IsThumb; } + bool isThumb1() const { return IsThumb && (ThumbMode == Thumb1); } bool isThumb2() const { return IsThumb && (ThumbMode >= Thumb2); } bool useThumbBacktraces() const { return UseThumbBacktraces; } diff --git a/lib/Target/ARM/ARMTargetMachine.cpp b/lib/Target/ARM/ARMTargetMachine.cpp index 1dc7d19aa10..7033907bbb0 100644 --- a/lib/Target/ARM/ARMTargetMachine.cpp +++ b/lib/Target/ARM/ARMTargetMachine.cpp @@ -23,6 +23,9 @@ #include "llvm/Target/TargetOptions.h" using namespace llvm; +static cl::opt +EnablePreLdStOpti("arm-pre-alloc-loadstore-opti", cl::Hidden, + cl::desc("Enable pre-regalloc load store optimization pass")); static cl::opt DisableLdStOpti("disable-arm-loadstore-opti", cl::Hidden, cl::desc("Disable load store optimization pass")); static cl::opt DisableIfConversion("disable-arm-if-conversion",cl::Hidden, @@ -144,6 +147,16 @@ bool ARMTargetMachine::addInstSelector(PassManagerBase &PM, return false; } +bool ARMTargetMachine::addPreRegAlloc(PassManagerBase &PM, + CodeGenOpt::Level OptLevel) { + if (!EnablePreLdStOpti) + return false; + // FIXME: temporarily disabling load / store optimization pass for Thumb mode. + if (OptLevel != CodeGenOpt::None && !DisableLdStOpti && !Subtarget.isThumb()) + PM.add(createARMLoadStoreOptimizationPass(true)); + return true; +} + bool ARMTargetMachine::addPreEmitPass(PassManagerBase &PM, CodeGenOpt::Level OptLevel) { // FIXME: temporarily disabling load / store optimization pass for Thumb mode. diff --git a/lib/Target/ARM/ARMTargetMachine.h b/lib/Target/ARM/ARMTargetMachine.h index 916a8aa9e98..7192c1bb618 100644 --- a/lib/Target/ARM/ARMTargetMachine.h +++ b/lib/Target/ARM/ARMTargetMachine.h @@ -71,6 +71,7 @@ public: // Pass Pipeline Configuration virtual bool addInstSelector(PassManagerBase &PM, CodeGenOpt::Level OptLevel); + virtual bool addPreRegAlloc(PassManagerBase &PM, CodeGenOpt::Level OptLevel); virtual bool addPreEmitPass(PassManagerBase &PM, CodeGenOpt::Level OptLevel); virtual bool addAssemblyEmitter(PassManagerBase &PM, CodeGenOpt::Level OptLevel, diff --git a/lib/Target/CMakeLists.txt b/lib/Target/CMakeLists.txt index 1cf0a91078f..7cffd0e53c1 100644 --- a/lib/Target/CMakeLists.txt +++ b/lib/Target/CMakeLists.txt @@ -5,6 +5,7 @@ add_llvm_library(LLVMTarget Target.cpp TargetAsmInfo.cpp TargetData.cpp + TargetELFWriterInfo.cpp TargetFrameInfo.cpp TargetInstrInfo.cpp TargetMachOWriterInfo.cpp @@ -14,4 +15,4 @@ add_llvm_library(LLVMTarget TargetSubtarget.cpp ) -# TODO: Support other targets besides X86. See Makefile. \ No newline at end of file +# TODO: Support other targets besides X86. See Makefile. diff --git a/lib/Target/PIC16/PIC16AsmPrinter.cpp b/lib/Target/PIC16/PIC16AsmPrinter.cpp index b42ee4524be..f9a88012650 100644 --- a/lib/Target/PIC16/PIC16AsmPrinter.cpp +++ b/lib/Target/PIC16/PIC16AsmPrinter.cpp @@ -33,8 +33,9 @@ bool PIC16AsmPrinter::printMachineInstruction(const MachineInstr *MI) { return true; } -/// runOnMachineFunction - This uses the printInstruction() -/// method to print assembly for each instruction. +/// runOnMachineFunction - This emits the frame section, autos section and +/// assembly for each instruction. Also takes care of function begin debug +/// directive and file begin debug directive (if required) for the function. /// bool PIC16AsmPrinter::runOnMachineFunction(MachineFunction &MF) { this->MF = &MF; @@ -47,20 +48,38 @@ bool PIC16AsmPrinter::runOnMachineFunction(MachineFunction &MF) { const Function *F = MF.getFunction(); CurrentFnName = Mang->getValueName(F); - DbgInfo.EmitFileDirective(F); - // Emit the function variables. + // Iterate over the first basic block instructions to find if it has a + // DebugLoc. If so emit .file directive. Instructions such as movlw do not + // have valid DebugLoc, so need to iterate over instructions. + MachineFunction::const_iterator I = MF.begin(); + for (MachineBasicBlock::const_iterator MBBI = I->begin(), E = I->end(); + MBBI != E; MBBI++) { + const DebugLoc DLoc = MBBI->getDebugLoc(); + if (!DLoc.isUnknown()) { + GlobalVariable *CU = MF.getDebugLocTuple(DLoc).CompileUnit; + unsigned line = MF.getDebugLocTuple(DLoc).Line; + DbgInfo.EmitFileDirective(CU); + DbgInfo.SetFunctBeginLine(line); + break; + } + } + + // Emit the function frame (args and temps). EmitFunctionFrame(MF); - // Emit function begin debug directives + // Emit function begin debug directive. DbgInfo.EmitFunctBeginDI(F); + // Emit the autos section of function. EmitAutos(CurrentFnName); + + // Now emit the instructions of function in its code section. const char *codeSection = PAN::getCodeSectionName(CurrentFnName).c_str(); const Section *fCodeSection = TAI->getNamedSection(codeSection, SectionFlags::Code); - O << "\n"; // Start the Code Section. + O << "\n"; SwitchToSection (fCodeSection); // Emit the frame address of the function at the beginning of code. @@ -77,14 +96,17 @@ bool PIC16AsmPrinter::runOnMachineFunction(MachineFunction &MF) { // Print out code for the function. for (MachineFunction::const_iterator I = MF.begin(), E = MF.end(); I != E; ++I) { + // Print a label for the basic block. if (I != MF.begin()) { printBasicBlockLabel(I, true); O << '\n'; } + // Print a basic block. for (MachineBasicBlock::const_iterator II = I->begin(), E = I->end(); II != E; ++II) { + // Emit the line directive if source line changed. const DebugLoc DL = II->getDebugLoc(); if (!DL.isUnknown()) { @@ -102,6 +124,7 @@ bool PIC16AsmPrinter::runOnMachineFunction(MachineFunction &MF) { // Emit function end debug directives. DbgInfo.EmitFunctEndDI(F, CurLine); + return false; // we didn't modify anything. } @@ -158,11 +181,16 @@ void PIC16AsmPrinter::printOperand(const MachineInstr *MI, int opNum) { } } +/// printCCOperand - Print the cond code operand. +/// void PIC16AsmPrinter::printCCOperand(const MachineInstr *MI, int opNum) { int CC = (int)MI->getOperand(opNum).getImm(); O << PIC16CondCodeToString((PIC16CC::CondCodes)CC); } +/// printLibcallDecls - print the extern declarations for compiler +/// intrinsics. +/// void PIC16AsmPrinter::printLibcallDecls(void) { // If no libcalls used, return. if (LibcallDecls.empty()) return; @@ -180,6 +208,10 @@ void PIC16AsmPrinter::printLibcallDecls(void) { O << TAI->getCommentString() << "External decls for libcalls - END." <<"\n"; } +/// doInitialization - Perfrom Module level initializations here. +/// One task that we do here is to sectionize all global variables. +/// The MemSelOptimizer pass depends on the sectionizing. +/// bool PIC16AsmPrinter::doInitialization (Module &M) { bool Result = AsmPrinter::doInitialization(M); @@ -194,23 +226,23 @@ bool PIC16AsmPrinter::doInitialization (Module &M) { I->setSection(TAI->SectionForGlobal(I)->getName()); } - DbgInfo.EmitFileDirective(M); + DbgInfo.Init(M); EmitFunctionDecls(M); EmitUndefinedVars(M); EmitDefinedVars(M); EmitIData(M); EmitUData(M); EmitRomData(M); - DbgInfo.PopulateFunctsDI(M); return Result; } -// Emit extern decls for functions imported from other modules, and emit -// global declarations for function defined in this module and which are -// available to other modules. +/// Emit extern decls for functions imported from other modules, and emit +/// global declarations for function defined in this module and which are +/// available to other modules. +/// void PIC16AsmPrinter::EmitFunctionDecls (Module &M) { // Emit declarations for external functions. - O << TAI->getCommentString() << "Function Declarations - BEGIN." <<"\n"; + O <<"\n"<getCommentString() << "Function Declarations - BEGIN." <<"\n"; for (Module::iterator I = M.begin(), E = M.end(); I != E; I++) { std::string Name = Mang->getValueName(I); if (Name.compare("@abort") == 0) @@ -280,6 +312,7 @@ void PIC16AsmPrinter::EmitRomData (Module &M) bool PIC16AsmPrinter::doFinalization(Module &M) { printLibcallDecls(); + EmitRemainingAutos(); DbgInfo.EmitVarDebugInfo(M); DbgInfo.EmitEOF(); O << "\n\t" << "END\n"; @@ -383,6 +416,8 @@ void PIC16AsmPrinter::EmitAutos (std::string FunctName) for (unsigned i = 0; i < AutosSections.size(); i++) { O << "\n"; if (AutosSections[i]->S_->getName() == SectionName) { + // Set the printing status to true + AutosSections[i]->setPrintedStatus(true); SwitchToSection(AutosSections[i]->S_); std::vector Items = AutosSections[i]->Items; for (unsigned j = 0; j < Items.size(); j++) { @@ -398,3 +433,34 @@ void PIC16AsmPrinter::EmitAutos (std::string FunctName) } } +// Print autos that were not printed during the code printing of functions. +// As the functions might themselves would have got deleted by the optimizer. +void PIC16AsmPrinter::EmitRemainingAutos() +{ + const TargetData *TD = TM.getTargetData(); + + // Now print Autos section for this function. + std::vector AutosSections = PTAI->AutosSections; + for (unsigned i = 0; i < AutosSections.size(); i++) { + + // if the section is already printed then don't print again + if (AutosSections[i]->isPrinted()) + continue; + + // Set status as printed + AutosSections[i]->setPrintedStatus(true); + + O << "\n"; + SwitchToSection(AutosSections[i]->S_); + std::vector Items = AutosSections[i]->Items; + for (unsigned j = 0; j < Items.size(); j++) { + std::string VarName = Mang->getValueName(Items[j]); + Constant *C = Items[j]->getInitializer(); + const Type *Ty = C->getType(); + unsigned Size = TD->getTypeAllocSize(Ty); + // Emit memory reserve directive. + O << VarName << " RES " << Size << "\n"; + } + } +} + diff --git a/lib/Target/PIC16/PIC16AsmPrinter.h b/lib/Target/PIC16/PIC16AsmPrinter.h index 2545dfd680a..8bdcf7258ef 100644 --- a/lib/Target/PIC16/PIC16AsmPrinter.h +++ b/lib/Target/PIC16/PIC16AsmPrinter.h @@ -52,6 +52,7 @@ namespace llvm { void EmitIData (Module &M); void EmitUData (Module &M); void EmitAutos (std::string FunctName); + void EmitRemainingAutos (); void EmitRomData (Module &M); void EmitFunctionFrame(MachineFunction &MF); void printLibcallDecls(void); diff --git a/lib/Target/PIC16/PIC16DebugInfo.cpp b/lib/Target/PIC16/PIC16DebugInfo.cpp index faf4590b43e..d7ebea7fa9d 100644 --- a/lib/Target/PIC16/PIC16DebugInfo.cpp +++ b/lib/Target/PIC16/PIC16DebugInfo.cpp @@ -18,13 +18,6 @@ using namespace llvm; -PIC16DbgInfo::~PIC16DbgInfo() { - for(std::map::iterator i = FunctNameMap.begin(); - i!=FunctNameMap.end(); i++) - delete i->second; - FunctNameMap.clear(); -} - void PIC16DbgInfo::PopulateDebugInfo(DIType Ty, unsigned short &TypeNo, bool &HasAux, int Aux[], std::string &TypeName) { @@ -70,7 +63,7 @@ void PIC16DbgInfo::PopulateDebugInfo(DIType Ty, unsigned short &TypeNo, } HasAux = true; // In auxillary entry for array, 7th and 8th byte represent array size. - Aux[6] = size; + Aux[6] = size & 0xff; Aux[7] = size >> 8; DIType BaseType = CTy.getTypeDerivedFrom(); PopulateDebugInfo(BaseType, TypeNo, HasAux, Aux, TypeName); @@ -86,10 +79,14 @@ void PIC16DbgInfo::PopulateDebugInfo(DIType Ty, unsigned short &TypeNo, else TypeNo = TypeNo | PIC16Dbg::T_UNION; CTy.getName(TypeName); - unsigned size = CTy.getSizeInBits()/8; + // UniqueSuffix is .number where number is obtained from + // llvm.dbg.composite. + std::string UniqueSuffix = "." + Ty.getGV()->getName().substr(18); + TypeName += UniqueSuffix; + unsigned short size = CTy.getSizeInBits()/8; // 7th and 8th byte represent size. HasAux = true; - Aux[6] = size; + Aux[6] = size & 0xff; Aux[7] = size >> 8; break; } @@ -145,37 +142,84 @@ short PIC16DbgInfo::getClass(DIGlobalVariable DIGV) { return ClassNo; } -void PIC16DbgInfo::PopulateFunctsDI(Module &M) { - GlobalVariable *Root = M.getGlobalVariable("llvm.dbg.subprograms"); - if (!Root) - return; - Constant *RootC = cast(*Root->use_begin()); - - for (Value::use_iterator UI = RootC->use_begin(), UE = Root->use_end(); - UI != UE; ++UI) - for (Value::use_iterator UUI = UI->use_begin(), UUE = UI->use_end(); - UUI != UUE; ++UUI) { - GlobalVariable *GVSP = cast(*UUI); - DISubprogram *SP = new DISubprogram(GVSP); - std::string Name; - SP->getLinkageName(Name); - FunctNameMap[Name] = SP; - } - return; +void PIC16DbgInfo::Init(Module &M) { + // Do all debug related initializations here. + EmitFileDirective(M); + EmitCompositeTypeDecls(M); } -DISubprogram* PIC16DbgInfo::getFunctDI(std::string FunctName) { - return FunctNameMap[FunctName]; +void PIC16DbgInfo::EmitCompositeTypeDecls(Module &M) { + for(iplist::iterator I = M.getGlobalList().begin(), + E = M.getGlobalList().end(); I != E; I++) { + // Structures and union declaration's debug info has llvm.dbg.composite + // in its name. + if(I->getName().find("llvm.dbg.composite") != std::string::npos) { + GlobalVariable *GV = cast(I); + DICompositeType CTy(GV); + if (CTy.getTag() == dwarf::DW_TAG_union_type || + CTy.getTag() == dwarf::DW_TAG_structure_type ) { + std::string name; + CTy.getName(name); + std::string DIVar = I->getName(); + // Get the number after llvm.dbg.composite and make UniqueSuffix from + // it. + std::string UniqueSuffix = "." + DIVar.substr(18); + std::string MangledCTyName = name + UniqueSuffix; + unsigned short size = CTy.getSizeInBits()/8; + int Aux[PIC16Dbg::AuxSize] = {0}; + // 7th and 8th byte represent size of structure/union. + Aux[6] = size & 0xff; + Aux[7] = size >> 8; + // Emit .def for structure/union tag. + if( CTy.getTag() == dwarf::DW_TAG_union_type) + EmitSymbol(MangledCTyName, PIC16Dbg::C_UNTAG); + else if (CTy.getTag() == dwarf::DW_TAG_structure_type) + EmitSymbol(MangledCTyName, PIC16Dbg::C_STRTAG); + + // Emit auxiliary debug information for structure/union tag. + EmitAuxEntry(MangledCTyName, Aux, PIC16Dbg::AuxSize); + unsigned long Value = 0; + DIArray Elements = CTy.getTypeArray(); + for (unsigned i = 0, N = Elements.getNumElements(); i < N; i++) { + DIDescriptor Element = Elements.getElement(i); + unsigned short TypeNo = 0; + bool HasAux = false; + int ElementAux[PIC16Dbg::AuxSize] = { 0 }; + std::string TypeName = ""; + std::string ElementName; + GlobalVariable *GV = Element.getGV(); + DIDerivedType DITy(GV); + DITy.getName(ElementName); + unsigned short ElementSize = DITy.getSizeInBits()/8; + // Get mangleddd name for this structure/union element. + std::string MangMemName = ElementName + UniqueSuffix; + PopulateDebugInfo(DITy, TypeNo, HasAux, ElementAux, TypeName); + short Class; + if( CTy.getTag() == dwarf::DW_TAG_union_type) + Class = PIC16Dbg::C_MOU; + else if (CTy.getTag() == dwarf::DW_TAG_structure_type) + Class = PIC16Dbg::C_MOS; + EmitSymbol(MangMemName, Class, TypeNo, Value); + if (CTy.getTag() == dwarf::DW_TAG_structure_type) + Value += ElementSize; + if (HasAux) + EmitAuxEntry(MangMemName, ElementAux, PIC16Dbg::AuxSize, TypeName); + } + // Emit mangled Symbol for end of structure/union. + std::string EOSSymbol = ".eos" + UniqueSuffix; + EmitSymbol(EOSSymbol, PIC16Dbg::C_EOS); + EmitAuxEntry(EOSSymbol, Aux, PIC16Dbg::AuxSize, MangledCTyName); + } + } + } } void PIC16DbgInfo::EmitFunctBeginDI(const Function *F) { std::string FunctName = F->getName(); - DISubprogram *SP = getFunctDI(FunctName); - if (SP) { + if (EmitDebugDirectives) { std::string FunctBeginSym = ".bf." + FunctName; std::string BlockBeginSym = ".bb." + FunctName; - int FunctBeginLine = SP->getLineNumber(); int BFAux[PIC16Dbg::AuxSize] = {0}; BFAux[4] = FunctBeginLine; BFAux[5] = FunctBeginLine >> 8; @@ -189,8 +233,7 @@ void PIC16DbgInfo::EmitFunctBeginDI(const Function *F) { void PIC16DbgInfo::EmitFunctEndDI(const Function *F, unsigned Line) { std::string FunctName = F->getName(); - DISubprogram *SP = getFunctDI(FunctName); - if (SP) { + if (EmitDebugDirectives) { std::string FunctEndSym = ".ef." + FunctName; std::string BlockEndSym = ".eb." + FunctName; @@ -208,14 +251,21 @@ void PIC16DbgInfo::EmitFunctEndDI(const Function *F, unsigned Line) { /// EmitAuxEntry - Emit Auxiliary debug information. /// -void PIC16DbgInfo::EmitAuxEntry(const std::string VarName, int Aux[], int num) { +void PIC16DbgInfo::EmitAuxEntry(const std::string VarName, int Aux[], int num, + std::string tag) { O << "\n\t.dim " << VarName << ", 1" ; + if (tag != "") + O << ", " << tag; for (int i = 0; igetName(); - DISubprogram *SP = getFunctDI(FunctName); - if (SP) { - std::string Dir, FN; - DICompileUnit CU = SP->getCompileUnit(); - std::string File = CU.getDirectory(Dir) + "/" + CU.getFilename(FN); - if ( File != CurFile) { +void PIC16DbgInfo::EmitFileDirective(GlobalVariable *CU, bool EmitEof) { + std::string Dir, FN; + DICompileUnit DIUnit(CU); + std::string File = DIUnit.getDirectory(Dir) + "/" + DIUnit.getFilename(FN); + if ( File != CurFile ) { + if (EmitEof) EmitEOF(); - O << "\n\t.file\t\"" << File << "\"\n" ; - CurFile = File; - } + O << "\n\t.file\t\"" << File << "\"\n" ; + CurFile = File; } } @@ -290,3 +324,6 @@ void PIC16DbgInfo::EmitEOF() { O << "\n\t.EOF"; } +void PIC16DbgInfo::SetFunctBeginLine(unsigned line) { + FunctBeginLine = line; +} diff --git a/lib/Target/PIC16/PIC16DebugInfo.h b/lib/Target/PIC16/PIC16DebugInfo.h index be3939336b2..9d503807b20 100644 --- a/lib/Target/PIC16/PIC16DebugInfo.h +++ b/lib/Target/PIC16/PIC16DebugInfo.h @@ -91,29 +91,36 @@ namespace llvm { class raw_ostream; class PIC16DbgInfo { - std::map FunctNameMap; raw_ostream &O; const TargetAsmInfo *TAI; std::string CurFile; + // EmitDebugDirectives is set if debug information is available. Default + // value for it is false. + bool EmitDebugDirectives; + unsigned FunctBeginLine; public: PIC16DbgInfo(raw_ostream &o, const TargetAsmInfo *T) : O(o), TAI(T) { - CurFile = ""; + CurFile = ""; + EmitDebugDirectives = false; } - ~PIC16DbgInfo(); void PopulateDebugInfo(DIType Ty, unsigned short &TypeNo, bool &HasAux, int Aux[], std::string &TypeName); unsigned GetTypeDebugNumber(std::string &type); short getClass(DIGlobalVariable DIGV); - void PopulateFunctsDI(Module &M); - DISubprogram *getFunctDI(std::string FunctName); void EmitFunctBeginDI(const Function *F); + void Init(Module &M); + void EmitCompositeTypeDecls(Module &M); void EmitFunctEndDI(const Function *F, unsigned Line); - void EmitAuxEntry(const std::string VarName, int Aux[], int num); - inline void EmitSymbol(std::string Name, int Class); + void EmitAuxEntry(const std::string VarName, int Aux[], + int num = PIC16Dbg::AuxSize, std::string tag = ""); + inline void EmitSymbol(std::string Name, short Class, + unsigned short Type = PIC16Dbg::T_NULL, + unsigned long Value = 0); void EmitVarDebugInfo(Module &M); void EmitFileDirective(Module &M); - void EmitFileDirective(const Function *F); + void EmitFileDirective(GlobalVariable *CU, bool EmitEof = true); void EmitEOF(); + void SetFunctBeginLine(unsigned line); }; } // end namespace llvm; #endif diff --git a/lib/Target/PIC16/PIC16ISelLowering.cpp b/lib/Target/PIC16/PIC16ISelLowering.cpp index ac9a143bc24..ba465f3fcc7 100644 --- a/lib/Target/PIC16/PIC16ISelLowering.cpp +++ b/lib/Target/PIC16/PIC16ISelLowering.cpp @@ -56,6 +56,17 @@ static const char *getIntrinsicName(unsigned opcode) { case RTLIB::SREM_I32: Basename = "srem.i32"; break; case RTLIB::UREM_I16: Basename = "urem.i16"; break; case RTLIB::UREM_I32: Basename = "urem.i32"; break; + + case RTLIB::FPTOSINT_F32_I32: + Basename = "f32_to_si32"; break; + case RTLIB::SINTTOFP_I32_F32: + Basename = "si32_to_f32"; break; + + case RTLIB::ADD_F32: Basename = "add.f32"; break; + case RTLIB::SUB_F32: Basename = "sub.f32"; break; + case RTLIB::MUL_F32: Basename = "mul.f32"; break; + case RTLIB::DIV_F32: Basename = "div.f32"; break; + } std::string prefix = PAN::getTagName(PAN::PREFIX_SYMBOL); @@ -113,7 +124,17 @@ PIC16TargetLowering::PIC16TargetLowering(PIC16TargetMachine &TM) // Unsigned remainder lib call names setLibcallName(RTLIB::UREM_I16, getIntrinsicName(RTLIB::UREM_I16)); setLibcallName(RTLIB::UREM_I32, getIntrinsicName(RTLIB::UREM_I32)); - + + // Floating point operations + setLibcallName(RTLIB::FPTOSINT_F32_I32, + getIntrinsicName(RTLIB::FPTOSINT_F32_I32)); + setLibcallName(RTLIB::SINTTOFP_I32_F32, + getIntrinsicName(RTLIB::SINTTOFP_I32_F32)); + setLibcallName(RTLIB::ADD_F32, getIntrinsicName(RTLIB::ADD_F32)); + setLibcallName(RTLIB::SUB_F32, getIntrinsicName(RTLIB::SUB_F32)); + setLibcallName(RTLIB::MUL_F32, getIntrinsicName(RTLIB::MUL_F32)); + setLibcallName(RTLIB::DIV_F32, getIntrinsicName(RTLIB::DIV_F32)); + setOperationAction(ISD::GlobalAddress, MVT::i16, Custom); setOperationAction(ISD::ExternalSymbol, MVT::i16, Custom); diff --git a/lib/Target/PIC16/PIC16TargetAsmInfo.h b/lib/Target/PIC16/PIC16TargetAsmInfo.h index e464e36f788..b7292b8ea78 100644 --- a/lib/Target/PIC16/PIC16TargetAsmInfo.h +++ b/lib/Target/PIC16/PIC16TargetAsmInfo.h @@ -33,9 +33,13 @@ namespace llvm { struct PIC16Section { const Section *S_; // Connection to actual Section. unsigned Size; // Total size of the objects contained. + bool SectionPrinted; std::vector Items; - PIC16Section (const Section *s) { S_ = s; Size = 0; } + PIC16Section (const Section *s) { S_ = s; Size = 0; + SectionPrinted = false;} + bool isPrinted() { return SectionPrinted ; } + void setPrintedStatus(bool status) { SectionPrinted = status ;} }; struct PIC16TargetAsmInfo : public TargetAsmInfo { diff --git a/lib/Target/TargetELFWriterInfo.cpp b/lib/Target/TargetELFWriterInfo.cpp new file mode 100644 index 00000000000..9651e65495e --- /dev/null +++ b/lib/Target/TargetELFWriterInfo.cpp @@ -0,0 +1,36 @@ +//===-- lib/Target/TargetELFWriterInfo.cpp - ELF Writer Info --0-*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file implements the TargetELFWriterInfo class. +// +//===----------------------------------------------------------------------===// + +#include "llvm/Function.h" +#include "llvm/Target/TargetELFWriterInfo.h" +#include "llvm/Target/TargetData.h" +#include "llvm/Target/TargetMachine.h" +using namespace llvm; + +TargetELFWriterInfo::TargetELFWriterInfo(TargetMachine &tm) : TM(tm) { + is64Bit = TM.getTargetData()->getPointerSizeInBits() == 64; + isLittleEndian = TM.getTargetData()->isLittleEndian(); +} + +TargetELFWriterInfo::~TargetELFWriterInfo() {} + +/// getFunctionAlignment - Returns the alignment for function 'F', targets +/// with different alignment constraints should overload this method +unsigned TargetELFWriterInfo::getFunctionAlignment(const Function *F) const { + const TargetData *TD = TM.getTargetData(); + unsigned FnAlign = F->getAlignment(); + unsigned TDAlign = TD->getPointerABIAlignment(); + unsigned Align = std::max(FnAlign, TDAlign); + assert(!(Align & (Align-1)) && "Alignment is not a power of two!"); + return Align; +} diff --git a/lib/Target/TargetMachine.cpp b/lib/Target/TargetMachine.cpp index dea293b502c..c487cb80530 100644 --- a/lib/Target/TargetMachine.cpp +++ b/lib/Target/TargetMachine.cpp @@ -30,6 +30,7 @@ namespace llvm { bool FiniteOnlyFPMathOption; bool HonorSignDependentRoundingFPMathOption; bool UseSoftFloat; + FloatABI::ABIType FloatABIType; bool NoImplicitFloat; bool NoZerosInBSS; bool ExceptionHandling; @@ -84,6 +85,19 @@ GenerateSoftFloatCalls("soft-float", cl::desc("Generate software floating point library calls"), cl::location(UseSoftFloat), cl::init(false)); +static cl::opt +FloatABIForCalls("float-abi", + cl::desc("Choose float ABI type"), + cl::location(FloatABIType), + cl::init(FloatABI::Default), + cl::values( + clEnumValN(FloatABI::Default, "default", + "Target default float ABI type"), + clEnumValN(FloatABI::Soft, "soft", + "Soft float ABI (implied by -soft-float)"), + clEnumValN(FloatABI::Hard, "hard", + "Hard float ABI (uses FP registers)"), + clEnumValEnd)); static cl::opt DontPlaceZerosInBSS("nozero-initialized-in-bss", cl::desc("Don't place zero-initialized symbols into bss section"), @@ -162,6 +176,14 @@ EnableStrongPHIElim(cl::Hidden, "strong-phi-elim", // TargetMachine Class // +TargetMachine::TargetMachine() + : AsmInfo(0) { + // Typically it will be subtargets that will adjust FloatABIType from Default + // to Soft or Hard. + if (UseSoftFloat) + FloatABIType = FloatABI::Soft; +} + TargetMachine::~TargetMachine() { delete AsmInfo; } diff --git a/lib/Target/X86/README.txt b/lib/Target/X86/README.txt index 710bd035743..3796aac57cb 100644 --- a/lib/Target/X86/README.txt +++ b/lib/Target/X86/README.txt @@ -482,35 +482,6 @@ _usesbb: //===---------------------------------------------------------------------===// -Currently we don't have elimination of redundant stack manipulations. Consider -the code: - -int %main() { -entry: - call fastcc void %test1( ) - call fastcc void %test2( sbyte* cast (void ()* %test1 to sbyte*) ) - ret int 0 -} - -declare fastcc void %test1() - -declare fastcc void %test2(sbyte*) - - -This currently compiles to: - - subl $16, %esp - call _test5 - addl $12, %esp - subl $16, %esp - movl $_test5, (%esp) - call _test6 - addl $12, %esp - -The add\sub pair is really unneeded here. - -//===---------------------------------------------------------------------===// - Consider the expansion of: define i32 @test3(i32 %X) { @@ -902,34 +873,6 @@ condition register is dead. xor reg reg is shorter than mov reg, #0. //===---------------------------------------------------------------------===// -We aren't matching RMW instructions aggressively -enough. Here's a reduced testcase (more in PR1160): - -define void @test(i32* %huge_ptr, i32* %target_ptr) { - %A = load i32* %huge_ptr ; [#uses=1] - %B = load i32* %target_ptr ; [#uses=1] - %C = or i32 %A, %B ; [#uses=1] - store i32 %C, i32* %target_ptr - ret void -} - -$ llvm-as < t.ll | llc -march=x86-64 - -_test: - movl (%rdi), %eax - orl (%rsi), %eax - movl %eax, (%rsi) - ret - -That should be something like: - -_test: - movl (%rdi), %eax - orl %eax, (%rsi) - ret - -//===---------------------------------------------------------------------===// - The following code: bb114.preheader: ; preds = %cond_next94 @@ -1897,3 +1840,60 @@ The second one is done for: Atom, Pentium Pro, all AMDs, Pentium 4, Nocona, Core 2, and "Generic" //===---------------------------------------------------------------------===// + +Testcase: +int a(int x) { return (x & 127) > 31; } + +Current output: + movl 4(%esp), %eax + andl $127, %eax + cmpl $31, %eax + seta %al + movzbl %al, %eax + ret + +Ideal output: + xorl %eax, %eax + testl $96, 4(%esp) + setne %al + ret + +We could do this transformation in instcombine, but it's only clearly +beneficial on platforms with a test instruction. + +//===---------------------------------------------------------------------===// +Testcase: +int x(int a) { return (a&0xf0)>>4; } + +Current output: + movl 4(%esp), %eax + shrl $4, %eax + andl $15, %eax + ret + +Ideal output: + movzbl 4(%esp), %eax + shrl $4, %eax + ret + +//===---------------------------------------------------------------------===// + +Testcase: +int x(int a) { return (a & 0x80) ? 0x100 : 0; } + +Current output: + testl $128, 4(%esp) + setne %al + movzbl %al, %eax + shll $8, %eax + ret + +Ideal output: + movl 4(%esp), %eax + addl %eax, %eax + andl $256, %eax + ret + +We generally want to fold shifted tests of a single bit into a shift+and on x86. + +//===---------------------------------------------------------------------===// diff --git a/lib/Target/X86/X86CallingConv.td b/lib/Target/X86/X86CallingConv.td index 7f99203a833..e9fcbd5a489 100644 --- a/lib/Target/X86/X86CallingConv.td +++ b/lib/Target/X86/X86CallingConv.td @@ -215,50 +215,6 @@ def CC_X86_Win64_C : CallingConv<[ CCIfType<[v8i8, v4i16, v2i32, v1i64], CCAssignToStack<8, 16>> ]>; -// Tail call convention (fast): One register is reserved for target address, -// namely R9 -def CC_X86_64_TailCall : CallingConv<[ - // Handles byval parameters. - CCIfByVal>, - - // Promote i8/i16 arguments to i32. - CCIfType<[i8, i16], CCPromoteToType>, - - // The 'nest' parameter, if any, is passed in R10. - CCIfNest>, - - // The first 6 integer arguments are passed in integer registers. - CCIfType<[i32], CCAssignToReg<[EDI, ESI, EDX, ECX, R8D]>>, - CCIfType<[i64], CCAssignToReg<[RDI, RSI, RDX, RCX, R8]>>, - - // The first 8 FP/Vector arguments are passed in XMM registers. - CCIfType<[f32, f64, v16i8, v8i16, v4i32, v2i64, v4f32, v2f64], - CCIfSubtarget<"hasSSE1()", - CCAssignToReg<[XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7]>>>, - - // The first 8 MMX (except for v1i64) vector arguments are passed in XMM - // registers on Darwin. - CCIfType<[v8i8, v4i16, v2i32, v2f32], - CCIfSubtarget<"isTargetDarwin()", - CCAssignToReg<[XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7]>>>, - - // The first 8 v1i64 vector arguments are passed in GPRs on Darwin. - CCIfType<[v1i64], - CCIfSubtarget<"isTargetDarwin()", - CCAssignToReg<[RDI, RSI, RDX, RCX, R8]>>>, - - // Integer/FP values get stored in stack slots that are 8 bytes in size and - // 8-byte aligned if there are no more registers to hold them. - CCIfType<[i32, i64, f32, f64], CCAssignToStack<8, 8>>, - - // Vectors get 16-byte stack slots that are 16-byte aligned. - CCIfType<[v16i8, v8i16, v4i32, v2i64, v4f32, v2f64], CCAssignToStack<16, 16>>, - - // __m64 vectors get 8-byte stack slots that are 8-byte aligned. - CCIfType<[v8i8, v4i16, v2i32, v1i64], CCAssignToStack<8, 8>> -]>; - - //===----------------------------------------------------------------------===// // X86 C Calling Convention //===----------------------------------------------------------------------===// diff --git a/lib/Target/X86/X86ELFWriterInfo.cpp b/lib/Target/X86/X86ELFWriterInfo.cpp index 260474160cd..d84034b9ed4 100644 --- a/lib/Target/X86/X86ELFWriterInfo.cpp +++ b/lib/Target/X86/X86ELFWriterInfo.cpp @@ -12,8 +12,27 @@ //===----------------------------------------------------------------------===// #include "X86ELFWriterInfo.h" +#include "llvm/Function.h" +#include "llvm/Target/TargetData.h" +#include "llvm/Target/TargetMachine.h" using namespace llvm; -X86ELFWriterInfo::X86ELFWriterInfo(bool is64Bit) : - TargetELFWriterInfo(is64Bit ? EM_X86_64 : EM_386) {} +X86ELFWriterInfo::X86ELFWriterInfo(TargetMachine &TM) + : TargetELFWriterInfo(TM) { + bool is64Bit = TM.getTargetData()->getPointerSizeInBits() == 64; + EMachine = is64Bit ? EM_X86_64 : EM_386; + } + X86ELFWriterInfo::~X86ELFWriterInfo() {} + +unsigned X86ELFWriterInfo::getFunctionAlignment(const Function *F) const { + unsigned FnAlign = 4; + + if (F->hasFnAttr(Attribute::OptimizeForSize)) + FnAlign = 1; + + if (F->getAlignment()) + FnAlign = Log2_32(F->getAlignment()); + + return (1 << FnAlign); +} diff --git a/lib/Target/X86/X86ELFWriterInfo.h b/lib/Target/X86/X86ELFWriterInfo.h index acfa50111d2..e9c5bc45cba 100644 --- a/lib/Target/X86/X86ELFWriterInfo.h +++ b/lib/Target/X86/X86ELFWriterInfo.h @@ -20,8 +20,10 @@ namespace llvm { class X86ELFWriterInfo : public TargetELFWriterInfo { public: - X86ELFWriterInfo(bool is64Bit); + X86ELFWriterInfo(TargetMachine &TM); virtual ~X86ELFWriterInfo(); + + virtual unsigned getFunctionAlignment(const Function *F) const; }; } // end llvm namespace diff --git a/lib/Target/X86/X86FastISel.cpp b/lib/Target/X86/X86FastISel.cpp index 33332e4cf74..2bcfd76a8cf 100644 --- a/lib/Target/X86/X86FastISel.cpp +++ b/lib/Target/X86/X86FastISel.cpp @@ -171,8 +171,6 @@ CCAssignFn *X86FastISel::CCAssignFnForCall(unsigned CC, bool isTaillCall) { if (Subtarget->is64Bit()) { if (Subtarget->isTargetWin64()) return CC_X86_Win64_C; - else if (CC == CallingConv::Fast && isTaillCall) - return CC_X86_64_TailCall; else return CC_X86_64_C; } diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp index 9e15a543bce..36e3ab2e82a 100644 --- a/lib/Target/X86/X86ISelLowering.cpp +++ b/lib/Target/X86/X86ISelLowering.cpp @@ -944,7 +944,7 @@ SDValue X86TargetLowering::LowerRET(SDValue Op, SelectionDAG &DAG) { SDValue StackAdjustment = TailCall.getOperand(2); assert(((TargetAddress.getOpcode() == ISD::Register && (cast(TargetAddress)->getReg() == X86::EAX || - cast(TargetAddress)->getReg() == X86::R9)) || + cast(TargetAddress)->getReg() == X86::R11)) || TargetAddress.getOpcode() == ISD::TargetExternalSymbol || TargetAddress.getOpcode() == ISD::TargetGlobalAddress) && "Expecting an global address, external symbol, or register"); @@ -1171,8 +1171,6 @@ CCAssignFn *X86TargetLowering::CCAssignFnForNode(unsigned CC) const { if (Subtarget->is64Bit()) { if (Subtarget->isTargetWin64()) return CC_X86_Win64_C; - else if (CC == CallingConv::Fast && PerformTailCallOpt) - return CC_X86_64_TailCall; else return CC_X86_64_C; } @@ -1799,7 +1797,7 @@ SDValue X86TargetLowering::LowerCALL(SDValue Op, SelectionDAG &DAG) { } else if (ExternalSymbolSDNode *S = dyn_cast(Callee)) { Callee = DAG.getTargetExternalSymbol(S->getSymbol(), getPointerTy()); } else if (IsTailCall) { - unsigned Opc = Is64Bit ? X86::R9 : X86::EAX; + unsigned Opc = Is64Bit ? X86::R11 : X86::EAX; Chain = DAG.getCopyToReg(Chain, dl, DAG.getRegister(Opc, getPointerTy()), @@ -7696,7 +7694,7 @@ static bool EltsFromConsecutiveLoads(ShuffleVectorSDNode *N, unsigned NumElems, SelectionDAG &DAG, MachineFrameInfo *MFI, const TargetLowering &TLI) { LDBase = NULL; - LastLoadedElt = -1; + LastLoadedElt = -1U; for (unsigned i = 0; i < NumElems; ++i) { if (N->getMaskElt(i) < 0) { if (!LDBase) diff --git a/lib/Target/X86/X86RegisterInfo.cpp b/lib/Target/X86/X86RegisterInfo.cpp index c733f26a8c8..6c0074e62f2 100644 --- a/lib/Target/X86/X86RegisterInfo.cpp +++ b/lib/Target/X86/X86RegisterInfo.cpp @@ -822,6 +822,13 @@ void X86RegisterInfo::emitPrologue(MachineFunction &MF) const { NumBytes = StackSize - X86FI->getCalleeSavedFrameSize(); } + unsigned ReadyLabelId = 0; + if (needsFrameMoves) { + // Mark effective beginning of when frame pointer is ready. + ReadyLabelId = MMI->NextLabelID(); + BuildMI(MBB, MBBI, DL, TII.get(X86::DBG_LABEL)).addImm(ReadyLabelId); + } + // Skip the callee-saved push instructions. while (MBBI != MBB.end() && (MBBI->getOpcode() == X86::PUSH32r || @@ -831,67 +838,61 @@ void X86RegisterInfo::emitPrologue(MachineFunction &MF) const { if (MBBI != MBB.end()) DL = MBBI->getDebugLoc(); - if (NumBytes) { // Adjust stack pointer: ESP -= numbytes. - if (NumBytes >= 4096 && Subtarget->isTargetCygMing()) { - // Check, whether EAX is livein for this function. - bool isEAXAlive = false; - for (MachineRegisterInfo::livein_iterator + // Adjust stack pointer: ESP -= numbytes. + if (NumBytes >= 4096 && Subtarget->isTargetCygMing()) { + // Check, whether EAX is livein for this function. + bool isEAXAlive = false; + for (MachineRegisterInfo::livein_iterator II = MF.getRegInfo().livein_begin(), EE = MF.getRegInfo().livein_end(); (II != EE) && !isEAXAlive; ++II) { - unsigned Reg = II->first; - isEAXAlive = (Reg == X86::EAX || Reg == X86::AX || - Reg == X86::AH || Reg == X86::AL); - } - - // Function prologue calls _alloca to probe the stack when allocating more - // than 4k bytes in one go. Touching the stack at 4K increments is - // necessary to ensure that the guard pages used by the OS virtual memory - // manager are allocated in correct sequence. - if (!isEAXAlive) { - BuildMI(MBB, MBBI, DL, TII.get(X86::MOV32ri), X86::EAX) - .addImm(NumBytes); - BuildMI(MBB, MBBI, DL, TII.get(X86::CALLpcrel32)) - .addExternalSymbol("_alloca"); - } else { - // Save EAX - BuildMI(MBB, MBBI, DL, TII.get(X86::PUSH32r)) - .addReg(X86::EAX, RegState::Kill); - - // Allocate NumBytes-4 bytes on stack. We'll also use 4 already - // allocated bytes for EAX. - BuildMI(MBB, MBBI, DL, TII.get(X86::MOV32ri), X86::EAX) - .addImm(NumBytes-4); - BuildMI(MBB, MBBI, DL, TII.get(X86::CALLpcrel32)) - .addExternalSymbol("_alloca"); - - // Restore EAX - MachineInstr *MI = addRegOffset(BuildMI(MF, DL, TII.get(X86::MOV32rm), - X86::EAX), - StackPtr, false, NumBytes-4); - MBB.insert(MBBI, MI); - } - } else { - // If there is an SUB32ri of ESP immediately before this instruction, - // merge the two. This can be the case when tail call elimination is - // enabled and the callee has more arguments then the caller. - NumBytes -= mergeSPUpdates(MBB, MBBI, StackPtr, true); - - // If there is an ADD32ri or SUB32ri of ESP immediately after this - // instruction, merge the two instructions. - mergeSPUpdatesDown(MBB, MBBI, StackPtr, &NumBytes); - - if (NumBytes) - emitSPUpdate(MBB, MBBI, StackPtr, -(int64_t)NumBytes, Is64Bit, TII); + unsigned Reg = II->first; + isEAXAlive = (Reg == X86::EAX || Reg == X86::AX || + Reg == X86::AH || Reg == X86::AL); } + + // Function prologue calls _alloca to probe the stack when allocating more + // than 4k bytes in one go. Touching the stack at 4K increments is necessary + // to ensure that the guard pages used by the OS virtual memory manager are + // allocated in correct sequence. + if (!isEAXAlive) { + BuildMI(MBB, MBBI, DL, TII.get(X86::MOV32ri), X86::EAX) + .addImm(NumBytes); + BuildMI(MBB, MBBI, DL, TII.get(X86::CALLpcrel32)) + .addExternalSymbol("_alloca"); + } else { + // Save EAX + BuildMI(MBB, MBBI, DL, TII.get(X86::PUSH32r)) + .addReg(X86::EAX, RegState::Kill); + + // Allocate NumBytes-4 bytes on stack. We'll also use 4 already + // allocated bytes for EAX. + BuildMI(MBB, MBBI, DL, TII.get(X86::MOV32ri), X86::EAX) + .addImm(NumBytes - 4); + BuildMI(MBB, MBBI, DL, TII.get(X86::CALLpcrel32)) + .addExternalSymbol("_alloca"); + + // Restore EAX + MachineInstr *MI = addRegOffset(BuildMI(MF, DL, TII.get(X86::MOV32rm), + X86::EAX), + StackPtr, false, NumBytes - 4); + MBB.insert(MBBI, MI); + } + } else if (NumBytes) { + // If there is an SUB32ri of ESP immediately before this instruction, merge + // the two. This can be the case when tail call elimination is enabled and + // the callee has more arguments then the caller. + NumBytes -= mergeSPUpdates(MBB, MBBI, StackPtr, true); + + // If there is an ADD32ri or SUB32ri of ESP immediately after this + // instruction, merge the two instructions. + mergeSPUpdatesDown(MBB, MBBI, StackPtr, &NumBytes); + + if (NumBytes) + emitSPUpdate(MBB, MBBI, StackPtr, -(int64_t)NumBytes, Is64Bit, TII); } - if (needsFrameMoves) { - // Mark effective beginning of when frame pointer is ready. - unsigned ReadyLabelId = 0; - ReadyLabelId = MMI->NextLabelID(); - BuildMI(MBB, MBBI, DL, TII.get(X86::DBG_LABEL)).addImm(ReadyLabelId); + if (needsFrameMoves) emitFrameMoves(MF, FrameLabelId, ReadyLabelId); - } } void X86RegisterInfo::emitEpilogue(MachineFunction &MF, diff --git a/lib/Target/X86/X86Subtarget.cpp b/lib/Target/X86/X86Subtarget.cpp index 03ce1aee0e8..56983ce7f16 100644 --- a/lib/Target/X86/X86Subtarget.cpp +++ b/lib/Target/X86/X86Subtarget.cpp @@ -350,6 +350,10 @@ X86Subtarget::X86Subtarget(const Module &M, const std::string &FS, bool is64Bit) , MaxInlineSizeThreshold(128) , Is64Bit(is64Bit) , TargetType(isELF) { // Default to ELF unless otherwise specified. + + // default to hard float ABI + if (FloatABIType == FloatABI::Default) + FloatABIType = FloatABI::Hard; // Determine default and user specified characteristics if (!FS.empty()) { diff --git a/lib/Target/X86/X86TargetMachine.cpp b/lib/Target/X86/X86TargetMachine.cpp index 88ab247b0d5..dfb055ff78c 100644 --- a/lib/Target/X86/X86TargetMachine.cpp +++ b/lib/Target/X86/X86TargetMachine.cpp @@ -133,8 +133,7 @@ X86TargetMachine::X86TargetMachine(const Module &M, const std::string &FS, DataLayout(Subtarget.getDataLayout()), FrameInfo(TargetFrameInfo::StackGrowsDown, Subtarget.getStackAlignment(), Subtarget.is64Bit() ? -8 : -4), - InstrInfo(*this), JITInfo(*this), TLInfo(*this), - ELFWriterInfo(Subtarget.is64Bit()) { + InstrInfo(*this), JITInfo(*this), TLInfo(*this), ELFWriterInfo(*this) { DefRelocModel = getRelocationModel(); // FIXME: Correctly select PIC model for Win64 stuff if (getRelocationModel() == Reloc::Default) { diff --git a/lib/Transforms/IPO/ArgumentPromotion.cpp b/lib/Transforms/IPO/ArgumentPromotion.cpp index 2bb6428060c..a6126340161 100644 --- a/lib/Transforms/IPO/ArgumentPromotion.cpp +++ b/lib/Transforms/IPO/ArgumentPromotion.cpp @@ -127,17 +127,8 @@ bool ArgPromotion::PromoteArguments(CallGraphNode *CGN) { // Second check: make sure that all callers are direct callers. We can't // transform functions that have indirect callers. - for (Value::use_iterator UI = F->use_begin(), E = F->use_end(); - UI != E; ++UI) { - CallSite CS = CallSite::get(*UI); - if (!CS.getInstruction()) // "Taking the address" of the function - return false; - - // Ensure that this call site is CALLING the function, not passing it as - // an argument. - if (!CS.isCallee(UI)) - return false; - } + if (F->hasAddressTaken()) + return false; // Check to see which arguments are promotable. If an argument is promotable, // add it to ArgsToPromote. diff --git a/lib/Transforms/IPO/DeadArgumentElimination.cpp b/lib/Transforms/IPO/DeadArgumentElimination.cpp index 666db7e8d74..e480dadca89 100644 --- a/lib/Transforms/IPO/DeadArgumentElimination.cpp +++ b/lib/Transforms/IPO/DeadArgumentElimination.cpp @@ -175,15 +175,8 @@ bool DAE::DeleteDeadVarargs(Function &Fn) { if (Fn.isDeclaration() || !Fn.hasLocalLinkage()) return false; // Ensure that the function is only directly called. - for (Value::use_iterator I = Fn.use_begin(), E = Fn.use_end(); I != E; ++I) { - // If this use is anything other than a call site, give up. - CallSite CS = CallSite::get(*I); - Instruction *TheCall = CS.getInstruction(); - if (!TheCall) return false; // Not a direct call site? - - // The addr of this function is passed to the call. - if (!CS.isCallee(I)) return false; - } + if (Fn.hasAddressTaken()) + return false; // Okay, we know we can transform this function if safe. Scan its body // looking for calls to llvm.vastart. diff --git a/lib/Transforms/IPO/GlobalDCE.cpp b/lib/Transforms/IPO/GlobalDCE.cpp index db378b0d0b2..9c652b996ae 100644 --- a/lib/Transforms/IPO/GlobalDCE.cpp +++ b/lib/Transforms/IPO/GlobalDCE.cpp @@ -47,7 +47,6 @@ namespace { void GlobalIsNeeded(GlobalValue *GV); void MarkUsedGlobalsAsNeeded(Constant *C); - bool SafeToDestroyConstant(Constant* C); bool RemoveUnusedGlobalValue(GlobalValue &GV); }; } @@ -211,17 +210,3 @@ bool GlobalDCE::RemoveUnusedGlobalValue(GlobalValue &GV) { GV.removeDeadConstantUsers(); return GV.use_empty(); } - -// SafeToDestroyConstant - It is safe to destroy a constant iff it is only used -// by constants itself. Note that constants cannot be cyclic, so this test is -// pretty easy to implement recursively. -// -bool GlobalDCE::SafeToDestroyConstant(Constant *C) { - for (Value::use_iterator I = C->use_begin(), E = C->use_end(); I != E; ++I) - if (Constant *User = dyn_cast(*I)) { - if (!SafeToDestroyConstant(User)) return false; - } else { - return false; - } - return true; -} diff --git a/lib/Transforms/IPO/GlobalOpt.cpp b/lib/Transforms/IPO/GlobalOpt.cpp index 5f12825f043..9a1b2941907 100644 --- a/lib/Transforms/IPO/GlobalOpt.cpp +++ b/lib/Transforms/IPO/GlobalOpt.cpp @@ -136,16 +136,16 @@ struct VISIBILITY_HIDDEN GlobalStatus { } -/// ConstantIsDead - Return true if the specified constant is (transitively) -/// dead. The constant may be used by other constants (e.g. constant arrays and -/// constant exprs) as long as they are dead, but it cannot be used by anything -/// else. -static bool ConstantIsDead(Constant *C) { +// SafeToDestroyConstant - It is safe to destroy a constant iff it is only used +// by constants itself. Note that constants cannot be cyclic, so this test is +// pretty easy to implement recursively. +// +static bool SafeToDestroyConstant(Constant *C) { if (isa(C)) return false; for (Value::use_iterator UI = C->use_begin(), E = C->use_end(); UI != E; ++UI) if (Constant *CU = dyn_cast(*UI)) { - if (!ConstantIsDead(CU)) return false; + if (!SafeToDestroyConstant(CU)) return false; } else return false; return true; @@ -233,7 +233,7 @@ static bool AnalyzeGlobal(Value *V, GlobalStatus &GS, } else if (Constant *C = dyn_cast(*UI)) { GS.HasNonInstructionUser = true; // We might have a dead and dangling constant hanging off of here. - if (!ConstantIsDead(C)) + if (!SafeToDestroyConstant(C)) return true; } else { GS.HasNonInstructionUser = true; @@ -338,7 +338,7 @@ static bool CleanupConstantGlobalUsers(Value *V, Constant *Init) { } else if (Constant *C = dyn_cast(U)) { // If we have a chain of dead constantexprs or other things dangling from // us, and if they are all dead, nuke them without remorse. - if (ConstantIsDead(C)) { + if (SafeToDestroyConstant(C)) { C->destroyConstant(); // This could have invalidated UI, start over from scratch. CleanupConstantGlobalUsers(V, Init); @@ -354,7 +354,7 @@ static bool CleanupConstantGlobalUsers(Value *V, Constant *Init) { static bool isSafeSROAElementUse(Value *V) { // We might have a dead and dangling constant hanging off of here. if (Constant *C = dyn_cast(V)) - return ConstantIsDead(C); + return SafeToDestroyConstant(C); Instruction *I = dyn_cast(V); if (!I) return false; @@ -1769,22 +1769,6 @@ bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV, return false; } -/// OnlyCalledDirectly - Return true if the specified function is only called -/// directly. In other words, its address is never taken. -static bool OnlyCalledDirectly(Function *F) { - for (Value::use_iterator UI = F->use_begin(), E = F->use_end(); UI != E;++UI){ - Instruction *User = dyn_cast(*UI); - if (!User) return false; - if (!isa(User) && !isa(User)) return false; - - // See if the function address is passed as an argument. - for (User::op_iterator i = User->op_begin() + 1, e = User->op_end(); - i != e; ++i) - if (*i == F) return false; - } - return true; -} - /// ChangeCalleesToFastCall - Walk all of the direct calls of the specified /// function, changing them to FastCC. static void ChangeCalleesToFastCall(Function *F) { @@ -1830,7 +1814,7 @@ bool GlobalOpt::OptimizeFunctions(Module &M) { ++NumFnDeleted; } else if (F->hasLocalLinkage()) { if (F->getCallingConv() == CallingConv::C && !F->isVarArg() && - OnlyCalledDirectly(F)) { + !F->hasAddressTaken()) { // If this function has C calling conventions, is not a varargs // function, and is only called directly, promote it to use the Fast // calling convention. @@ -1841,7 +1825,7 @@ bool GlobalOpt::OptimizeFunctions(Module &M) { } if (F->getAttributes().hasAttrSomewhere(Attribute::Nest) && - OnlyCalledDirectly(F)) { + !F->hasAddressTaken()) { // The function is not used by a trampoline intrinsic, so it is safe // to remove the 'nest' attribute. RemoveNestAttribute(F); diff --git a/lib/Transforms/IPO/MergeFunctions.cpp b/lib/Transforms/IPO/MergeFunctions.cpp index 17bc2d41a4c..5693cc0fc3b 100644 --- a/lib/Transforms/IPO/MergeFunctions.cpp +++ b/lib/Transforms/IPO/MergeFunctions.cpp @@ -9,10 +9,6 @@ // // This pass looks for equivalent functions that are mergable and folds them. // -// A Function will not be analyzed if: -// * it is overridable at runtime (except for weak linkage), or -// * it is used by anything other than the callee parameter of a call/invoke -// // A hash is computed from the function, based on its type and number of // basic blocks. // @@ -24,8 +20,6 @@ // When a match is found, the functions are folded. We can only fold two // functions when we know that the definition of one of them is not // overridable. -// * fold a function marked internal by replacing all of its users. -// * fold extern or weak functions by replacing them with a global alias // //===----------------------------------------------------------------------===// // @@ -48,6 +42,7 @@ #define DEBUG_TYPE "mergefunc" #include "llvm/Transforms/IPO.h" #include "llvm/ADT/DenseMap.h" +#include "llvm/ADT/FoldingSet.h" #include "llvm/ADT/Statistic.h" #include "llvm/Constants.h" #include "llvm/InlineAsm.h" @@ -62,7 +57,6 @@ using namespace llvm; STATISTIC(NumFunctionsMerged, "Number of functions merged"); -STATISTIC(NumMergeFails, "Number of identical function pairings not merged"); namespace { struct VISIBILITY_HIDDEN MergeFunctions : public ModulePass { @@ -81,16 +75,169 @@ ModulePass *llvm::createMergeFunctionsPass() { return new MergeFunctions(); } +// ===----------------------------------------------------------------------=== +// Comparison of functions +// ===----------------------------------------------------------------------=== + static unsigned long hash(const Function *F) { - return F->size() ^ reinterpret_cast(F->getType()); - //return F->size() ^ F->arg_size() ^ F->getReturnType(); + const FunctionType *FTy = F->getFunctionType(); + + FoldingSetNodeID ID; + ID.AddInteger(F->size()); + ID.AddInteger(F->getCallingConv()); + ID.AddBoolean(F->hasGC()); + ID.AddBoolean(FTy->isVarArg()); + ID.AddInteger(FTy->getReturnType()->getTypeID()); + for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i) + ID.AddInteger(FTy->getParamType(i)->getTypeID()); + return ID.ComputeHash(); +} + +/// IgnoreBitcasts - given a bitcast, returns the first non-bitcast found by +/// walking the chain of cast operands. Otherwise, returns the argument. +static Value* IgnoreBitcasts(Value *V) { + while (BitCastInst *BC = dyn_cast(V)) + V = BC->getOperand(0); + + return V; +} + +/// isEquivalentType - any two pointers are equivalent. Otherwise, standard +/// type equivalence rules apply. +static bool isEquivalentType(const Type *Ty1, const Type *Ty2) { + if (Ty1 == Ty2) + return true; + if (Ty1->getTypeID() != Ty2->getTypeID()) + return false; + + switch(Ty1->getTypeID()) { + case Type::VoidTyID: + case Type::FloatTyID: + case Type::DoubleTyID: + case Type::X86_FP80TyID: + case Type::FP128TyID: + case Type::PPC_FP128TyID: + case Type::LabelTyID: + case Type::MetadataTyID: + return true; + + case Type::IntegerTyID: + case Type::OpaqueTyID: + // Ty1 == Ty2 would have returned true earlier. + return false; + + default: + assert(0 && "Unknown type!"); + return false; + + case Type::PointerTyID: { + const PointerType *PTy1 = cast(Ty1); + const PointerType *PTy2 = cast(Ty2); + return PTy1->getAddressSpace() == PTy2->getAddressSpace(); + } + + case Type::StructTyID: { + const StructType *STy1 = cast(Ty1); + const StructType *STy2 = cast(Ty2); + if (STy1->getNumElements() != STy2->getNumElements()) + return false; + + if (STy1->isPacked() != STy2->isPacked()) + return false; + + for (unsigned i = 0, e = STy1->getNumElements(); i != e; ++i) { + if (!isEquivalentType(STy1->getElementType(i), STy2->getElementType(i))) + return false; + } + return true; + } + + case Type::FunctionTyID: { + const FunctionType *FTy1 = cast(Ty1); + const FunctionType *FTy2 = cast(Ty2); + if (FTy1->getNumParams() != FTy2->getNumParams() || + FTy1->isVarArg() != FTy2->isVarArg()) + return false; + + if (!isEquivalentType(FTy1->getReturnType(), FTy2->getReturnType())) + return false; + + for (unsigned i = 0, e = FTy1->getNumParams(); i != e; ++i) { + if (!isEquivalentType(FTy1->getParamType(i), FTy2->getParamType(i))) + return false; + } + return true; + } + + case Type::ArrayTyID: + case Type::VectorTyID: { + const SequentialType *STy1 = cast(Ty1); + const SequentialType *STy2 = cast(Ty2); + return isEquivalentType(STy1->getElementType(), STy2->getElementType()); + } + } +} + +/// isEquivalentOperation - determine whether the two operations are the same +/// except that pointer-to-A and pointer-to-B are equivalent. This should be +/// kept in sync with Instruction::isSameOperationAs. +static bool +isEquivalentOperation(const Instruction *I1, const Instruction *I2) { + if (I1->getOpcode() != I2->getOpcode() || + I1->getNumOperands() != I2->getNumOperands() || + !isEquivalentType(I1->getType(), I2->getType())) + return false; + + // We have two instructions of identical opcode and #operands. Check to see + // if all operands are the same type + for (unsigned i = 0, e = I1->getNumOperands(); i != e; ++i) + if (!isEquivalentType(I1->getOperand(i)->getType(), + I2->getOperand(i)->getType())) + return false; + + // Check special state that is a part of some instructions. + if (const LoadInst *LI = dyn_cast(I1)) + return LI->isVolatile() == cast(I2)->isVolatile() && + LI->getAlignment() == cast(I2)->getAlignment(); + if (const StoreInst *SI = dyn_cast(I1)) + return SI->isVolatile() == cast(I2)->isVolatile() && + SI->getAlignment() == cast(I2)->getAlignment(); + if (const CmpInst *CI = dyn_cast(I1)) + return CI->getPredicate() == cast(I2)->getPredicate(); + if (const CallInst *CI = dyn_cast(I1)) + return CI->isTailCall() == cast(I2)->isTailCall() && + CI->getCallingConv() == cast(I2)->getCallingConv() && + CI->getAttributes().getRawPointer() == + cast(I2)->getAttributes().getRawPointer(); + if (const InvokeInst *CI = dyn_cast(I1)) + return CI->getCallingConv() == cast(I2)->getCallingConv() && + CI->getAttributes().getRawPointer() == + cast(I2)->getAttributes().getRawPointer(); + if (const InsertValueInst *IVI = dyn_cast(I1)) { + if (IVI->getNumIndices() != cast(I2)->getNumIndices()) + return false; + for (unsigned i = 0, e = IVI->getNumIndices(); i != e; ++i) + if (IVI->idx_begin()[i] != cast(I2)->idx_begin()[i]) + return false; + return true; + } + if (const ExtractValueInst *EVI = dyn_cast(I1)) { + if (EVI->getNumIndices() != cast(I2)->getNumIndices()) + return false; + for (unsigned i = 0, e = EVI->getNumIndices(); i != e; ++i) + if (EVI->idx_begin()[i] != cast(I2)->idx_begin()[i]) + return false; + return true; + } + + return true; } static bool compare(const Value *V, const Value *U) { assert(!isa(V) && !isa(U) && "Must not compare basic blocks."); - assert(V->getType() == U->getType() && + assert(isEquivalentType(V->getType(), U->getType()) && "Two of the same operation have operands of different type."); // TODO: If the constant is an expression of F, we should accept that it's @@ -117,19 +264,39 @@ static bool compare(const Value *V, const Value *U) { static bool equals(const BasicBlock *BB1, const BasicBlock *BB2, DenseMap &ValueMap, DenseMap &SpeculationMap) { - // Specutively add it anyways. If it's false, we'll notice a difference later, and - // this won't matter. + // Speculatively add it anyways. If it's false, we'll notice a difference + // later, and this won't matter. ValueMap[BB1] = BB2; BasicBlock::const_iterator FI = BB1->begin(), FE = BB1->end(); BasicBlock::const_iterator GI = BB2->begin(), GE = BB2->end(); do { - if (!FI->isSameOperationAs(const_cast(&*GI))) + if (isa(FI)) { + ++FI; + continue; + } + if (isa(GI)) { + ++GI; + continue; + } + + if (!isEquivalentOperation(FI, GI)) return false; - if (FI->getNumOperands() != GI->getNumOperands()) - return false; + if (isa(FI)) { + const GetElementPtrInst *GEPF = cast(FI); + const GetElementPtrInst *GEPG = cast(GI); + if (GEPF->hasAllZeroIndices() && GEPG->hasAllZeroIndices()) { + // It's effectively a bitcast. + ++FI, ++GI; + continue; + } + + // TODO: we only really care about the elements before the index + if (FI->getOperand(0)->getType() != GI->getOperand(0)->getType()) + return false; + } if (ValueMap[FI] == GI) { ++FI, ++GI; @@ -140,8 +307,8 @@ static bool equals(const BasicBlock *BB1, const BasicBlock *BB2, return false; for (unsigned i = 0, e = FI->getNumOperands(); i != e; ++i) { - Value *OpF = FI->getOperand(i); - Value *OpG = GI->getOperand(i); + Value *OpF = IgnoreBitcasts(FI->getOperand(i)); + Value *OpG = IgnoreBitcasts(GI->getOperand(i)); if (ValueMap[OpF] == OpG) continue; @@ -149,10 +316,8 @@ static bool equals(const BasicBlock *BB1, const BasicBlock *BB2, if (ValueMap[OpF] != NULL) return false; - assert(OpF->getType() == OpG->getType() && - "Two of the same operation has operands of different type."); - - if (OpF->getValueID() != OpG->getValueID()) + if (OpF->getValueID() != OpG->getValueID() || + !isEquivalentType(OpF->getType(), OpG->getType())) return false; if (isa(FI)) { @@ -203,14 +368,15 @@ static bool equals(const Function *F, const Function *G) { if (F->hasSection() && F->getSection() != G->getSection()) return false; + if (F->isVarArg() != G->isVarArg()) + return false; + // TODO: if it's internal and only used in direct calls, we could handle this // case too. if (F->getCallingConv() != G->getCallingConv()) return false; - // TODO: We want to permit cases where two functions take T* and S* but - // only load or store them into T** and S**. - if (F->getType() != G->getType()) + if (!isEquivalentType(F->getFunctionType(), G->getFunctionType())) return false; DenseMap ValueMap; @@ -237,88 +403,212 @@ static bool equals(const Function *F, const Function *G) { return true; } -static bool fold(std::vector &FnVec, unsigned i, unsigned j) { - if (FnVec[i]->mayBeOverridden() && !FnVec[j]->mayBeOverridden()) - std::swap(FnVec[i], FnVec[j]); +// ===----------------------------------------------------------------------=== +// Folding of functions +// ===----------------------------------------------------------------------=== +// Cases: +// * F is external strong, G is external strong: +// turn G into a thunk to F (1) +// * F is external strong, G is external weak: +// turn G into a thunk to F (1) +// * F is external weak, G is external weak: +// unfoldable +// * F is external strong, G is internal: +// address of G taken: +// turn G into a thunk to F (1) +// address of G not taken: +// make G an alias to F (2) +// * F is internal, G is external weak +// address of F is taken: +// turn G into a thunk to F (1) +// address of F is not taken: +// make G an alias of F (2) +// * F is internal, G is internal: +// address of F and G are taken: +// turn G into a thunk to F (1) +// address of G is not taken: +// make G an alias to F (2) +// +// alias requires linkage == (external,local,weak) fallback to creating a thunk +// external means 'externally visible' linkage != (internal,private) +// internal means linkage == (internal,private) +// weak means linkage mayBeOverridable +// being external implies that the address is taken +// +// 1. turn G into a thunk to F +// 2. make G an alias to F + +enum LinkageCategory { + ExternalStrong, + ExternalWeak, + Internal +}; + +static LinkageCategory categorize(const Function *F) { + switch (F->getLinkage()) { + case GlobalValue::InternalLinkage: + case GlobalValue::PrivateLinkage: + return Internal; + + case GlobalValue::WeakAnyLinkage: + case GlobalValue::WeakODRLinkage: + case GlobalValue::ExternalWeakLinkage: + return ExternalWeak; + + case GlobalValue::ExternalLinkage: + case GlobalValue::AvailableExternallyLinkage: + case GlobalValue::LinkOnceAnyLinkage: + case GlobalValue::LinkOnceODRLinkage: + case GlobalValue::AppendingLinkage: + case GlobalValue::DLLImportLinkage: + case GlobalValue::DLLExportLinkage: + case GlobalValue::GhostLinkage: + case GlobalValue::CommonLinkage: + return ExternalStrong; + } + + assert(0 && "Unknown LinkageType."); + return ExternalWeak; +} + +static void ThunkGToF(Function *F, Function *G) { + Function *NewG = Function::Create(G->getFunctionType(), G->getLinkage(), "", + G->getParent()); + BasicBlock *BB = BasicBlock::Create("", NewG); + + std::vector Args; + unsigned i = 0; + const FunctionType *FFTy = F->getFunctionType(); + for (Function::arg_iterator AI = NewG->arg_begin(), AE = NewG->arg_end(); + AI != AE; ++AI) { + if (FFTy->getParamType(i) == AI->getType()) + Args.push_back(AI); + else { + Value *BCI = new BitCastInst(AI, FFTy->getParamType(i), "", BB); + Args.push_back(BCI); + } + ++i; + } + + CallInst *CI = CallInst::Create(F, Args.begin(), Args.end(), "", BB); + CI->setTailCall(); + CI->setCallingConv(F->getCallingConv()); + if (NewG->getReturnType() == Type::VoidTy) { + ReturnInst::Create(BB); + } else if (CI->getType() != NewG->getReturnType()) { + Value *BCI = new BitCastInst(CI, NewG->getReturnType(), "", BB); + ReturnInst::Create(BCI, BB); + } else { + ReturnInst::Create(CI, BB); + } + + NewG->copyAttributesFrom(G); + NewG->takeName(G); + G->replaceAllUsesWith(NewG); + G->eraseFromParent(); + + // TODO: look at direct callers to G and make them all direct callers to F. +} + +static void AliasGToF(Function *F, Function *G) { + if (!G->hasExternalLinkage() && !G->hasLocalLinkage() && !G->hasWeakLinkage()) + return ThunkGToF(F, G); + + GlobalAlias *GA = new GlobalAlias( + G->getType(), G->getLinkage(), "", + ConstantExpr::getBitCast(F, G->getType()), G->getParent()); + F->setAlignment(std::max(F->getAlignment(), G->getAlignment())); + GA->takeName(G); + GA->setVisibility(G->getVisibility()); + G->replaceAllUsesWith(GA); + G->eraseFromParent(); +} + +static bool fold(std::vector &FnVec, unsigned i, unsigned j) { Function *F = FnVec[i]; Function *G = FnVec[j]; - if (!F->mayBeOverridden()) { - if (G->hasLocalLinkage()) { - F->setAlignment(std::max(F->getAlignment(), G->getAlignment())); - G->replaceAllUsesWith(F); - G->eraseFromParent(); - ++NumFunctionsMerged; - return true; - } + LinkageCategory catF = categorize(F); + LinkageCategory catG = categorize(G); - if (G->hasExternalLinkage() || G->hasWeakLinkage()) { - GlobalAlias *GA = new GlobalAlias(G->getType(), G->getLinkage(), "", - F, G->getParent()); - F->setAlignment(std::max(F->getAlignment(), G->getAlignment())); - GA->takeName(G); - GA->setVisibility(G->getVisibility()); - G->replaceAllUsesWith(GA); - G->eraseFromParent(); - ++NumFunctionsMerged; - return true; - } + if (catF == ExternalWeak || (catF == Internal && catG == ExternalStrong)) { + std::swap(FnVec[i], FnVec[j]); + std::swap(F, G); + std::swap(catF, catG); } - if (F->hasWeakLinkage() && G->hasWeakLinkage()) { - GlobalAlias *GA_F = new GlobalAlias(F->getType(), F->getLinkage(), "", - 0, F->getParent()); - GA_F->takeName(F); - GA_F->setVisibility(F->getVisibility()); - F->setAlignment(std::max(F->getAlignment(), G->getAlignment())); - F->replaceAllUsesWith(GA_F); - F->setName("folded." + GA_F->getName()); - F->setLinkage(GlobalValue::ExternalLinkage); - GA_F->setAliasee(F); + switch (catF) { + case ExternalStrong: + switch (catG) { + case ExternalStrong: + case ExternalWeak: + ThunkGToF(F, G); + break; + case Internal: + if (G->hasAddressTaken()) + ThunkGToF(F, G); + else + AliasGToF(F, G); + break; + } + break; - GlobalAlias *GA_G = new GlobalAlias(G->getType(), G->getLinkage(), "", - F, G->getParent()); - GA_G->takeName(G); - GA_G->setVisibility(G->getVisibility()); - G->replaceAllUsesWith(GA_G); - G->eraseFromParent(); + case ExternalWeak: { + assert(catG == ExternalWeak); - ++NumFunctionsMerged; - return true; + // Make them both thunks to the same internal function. + F->setAlignment(std::max(F->getAlignment(), G->getAlignment())); + Function *H = Function::Create(F->getFunctionType(), F->getLinkage(), "", + F->getParent()); + H->copyAttributesFrom(F); + H->takeName(F); + F->replaceAllUsesWith(H); + + ThunkGToF(F, G); + ThunkGToF(F, H); + + F->setLinkage(GlobalValue::InternalLinkage); + } break; + + case Internal: + switch (catG) { + case ExternalStrong: + assert(0); + // fall-through + case ExternalWeak: + if (F->hasAddressTaken()) + ThunkGToF(F, G); + else + AliasGToF(F, G); + break; + case Internal: { + bool addrTakenF = F->hasAddressTaken(); + bool addrTakenG = G->hasAddressTaken(); + if (!addrTakenF && addrTakenG) { + std::swap(FnVec[i], FnVec[j]); + std::swap(F, G); + std::swap(addrTakenF, addrTakenG); + } + + if (addrTakenF && addrTakenG) { + ThunkGToF(F, G); + } else { + assert(!addrTakenG); + AliasGToF(F, G); + } + } break; + } + break; } - DOUT << "Failed on " << F->getName() << " and " << G->getName() << "\n"; - - ++NumMergeFails; - return false; + ++NumFunctionsMerged; + return true; } -static bool hasAddressTaken(User *U) { - for (User::use_iterator I = U->use_begin(), E = U->use_end(); I != E; ++I) { - User *Use = *I; - - // 'call (bitcast @F to ...)' happens a lot. - while (isa(Use) && Use->hasOneUse()) { - Use = *Use->use_begin(); - } - - if (isa(Use)) { - if (hasAddressTaken(Use)) - return true; - } - - if (!isa(Use) && !isa(Use)) - return true; - - // Make sure we aren't passing U as a parameter to call instead of the - // callee. - if (CallSite(cast(Use)).hasArgument(U)) - return true; - } - - return false; -} +// ===----------------------------------------------------------------------=== +// Pass definition +// ===----------------------------------------------------------------------=== bool MergeFunctions::runOnModule(Module &M) { bool Changed = false; @@ -329,25 +619,19 @@ bool MergeFunctions::runOnModule(Module &M) { if (F->isDeclaration() || F->isIntrinsic()) continue; - if (!F->hasLocalLinkage() && !F->hasExternalLinkage() && - !F->hasWeakLinkage()) - continue; - - if (hasAddressTaken(F)) - continue; - FnMap[hash(F)].push_back(F); } - // TODO: instead of running in a loop, we could also fold functions in callgraph - // order. Constructing the CFG probably isn't cheaper than just running in a loop. + // TODO: instead of running in a loop, we could also fold functions in + // callgraph order. Constructing the CFG probably isn't cheaper than just + // running in a loop, unless it happened to already be available. bool LocalChanged; do { LocalChanged = false; + DOUT << "size: " << FnMap.size() << "\n"; for (std::map >::iterator I = FnMap.begin(), E = FnMap.end(); I != E; ++I) { - DOUT << "size: " << FnMap.size() << "\n"; std::vector &FnVec = I->second; DOUT << "hash (" << I->first << "): " << FnVec.size() << "\n"; diff --git a/lib/Transforms/IPO/PartialInlining.cpp b/lib/Transforms/IPO/PartialInlining.cpp new file mode 100644 index 00000000000..b3a2554039a --- /dev/null +++ b/lib/Transforms/IPO/PartialInlining.cpp @@ -0,0 +1,171 @@ +//===- PartialInlining.cpp - Inline parts of functions --------------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This pass performs partial inlining, typically by inlining an if statement +// that surrounds the body of the function. +// +//===----------------------------------------------------------------------===// + +#define DEBUG_TYPE "partialinlining" +#include "llvm/Transforms/IPO.h" +#include "llvm/Instructions.h" +#include "llvm/Module.h" +#include "llvm/Pass.h" +#include "llvm/Analysis/Dominators.h" +#include "llvm/Transforms/Utils/Cloning.h" +#include "llvm/Transforms/Utils/FunctionUtils.h" +#include "llvm/Support/Compiler.h" +#include "llvm/Support/CFG.h" +using namespace llvm; + +namespace { + struct VISIBILITY_HIDDEN PartialInliner : public ModulePass { + virtual void getAnalysisUsage(AnalysisUsage &AU) const { } + static char ID; // Pass identification, replacement for typeid + PartialInliner() : ModulePass(&ID) {} + + bool runOnModule(Module& M); + + private: + Function* unswitchFunction(Function* F); + }; +} + +char PartialInliner::ID = 0; +static RegisterPass X("partial-inliner", "Partial Inliner"); + +ModulePass* llvm::createPartialInliningPass() { return new PartialInliner(); } + +Function* PartialInliner::unswitchFunction(Function* F) { + // First, verify that this function is an unswitching candidate... + BasicBlock* entryBlock = F->begin(); + if (!isa(entryBlock->getTerminator())) + return 0; + + BasicBlock* returnBlock = 0; + BasicBlock* nonReturnBlock = 0; + unsigned returnCount = 0; + for (succ_iterator SI = succ_begin(entryBlock), SE = succ_end(entryBlock); + SI != SE; ++SI) + if (isa((*SI)->getTerminator())) { + returnBlock = *SI; + returnCount++; + } else + nonReturnBlock = *SI; + + if (returnCount != 1) + return 0; + + // Clone the function, so that we can hack away on it. + DenseMap ValueMap; + Function* duplicateFunction = CloneFunction(F, ValueMap); + duplicateFunction->setLinkage(GlobalValue::InternalLinkage); + F->getParent()->getFunctionList().push_back(duplicateFunction); + BasicBlock* newEntryBlock = cast(ValueMap[entryBlock]); + BasicBlock* newReturnBlock = cast(ValueMap[returnBlock]); + BasicBlock* newNonReturnBlock = cast(ValueMap[nonReturnBlock]); + + // Go ahead and update all uses to the duplicate, so that we can just + // use the inliner functionality when we're done hacking. + F->replaceAllUsesWith(duplicateFunction); + + // Special hackery is needed with PHI nodes that have inputs from more than + // one extracted block. For simplicity, just split the PHIs into a two-level + // sequence of PHIs, some of which will go in the extracted region, and some + // of which will go outside. + BasicBlock* preReturn = newReturnBlock; + newReturnBlock = newReturnBlock->splitBasicBlock( + newReturnBlock->getFirstNonPHI()); + BasicBlock::iterator I = preReturn->begin(); + BasicBlock::iterator Ins = newReturnBlock->begin(); + while (I != preReturn->end()) { + PHINode* OldPhi = dyn_cast(I); + if (!OldPhi) break; + + PHINode* retPhi = PHINode::Create(OldPhi->getType(), "", Ins); + OldPhi->replaceAllUsesWith(retPhi); + Ins = newReturnBlock->getFirstNonPHI(); + + retPhi->addIncoming(I, preReturn); + retPhi->addIncoming(OldPhi->getIncomingValueForBlock(newEntryBlock), + newEntryBlock); + OldPhi->removeIncomingValue(newEntryBlock); + + ++I; + } + newEntryBlock->getTerminator()->replaceUsesOfWith(preReturn, newReturnBlock); + + // Gather up the blocks that we're going to extract. + std::vector toExtract; + toExtract.push_back(newNonReturnBlock); + for (Function::iterator FI = duplicateFunction->begin(), + FE = duplicateFunction->end(); FI != FE; ++FI) + if (&*FI != newEntryBlock && &*FI != newReturnBlock && + &*FI != newNonReturnBlock) + toExtract.push_back(FI); + + // The CodeExtractor needs a dominator tree. + DominatorTree DT; + DT.runOnFunction(*duplicateFunction); + + // Extract the body of the the if. + Function* extractedFunction = ExtractCodeRegion(DT, toExtract); + + // Inline the top-level if test into all callers. + std::vector Users(duplicateFunction->use_begin(), + duplicateFunction->use_end()); + for (std::vector::iterator UI = Users.begin(), UE = Users.end(); + UI != UE; ++UI) + if (CallInst* CI = dyn_cast(*UI)) + InlineFunction(CI); + else if (InvokeInst* II = dyn_cast(*UI)) + InlineFunction(II); + + // Ditch the duplicate, since we're done with it, and rewrite all remaining + // users (function pointers, etc.) back to the original function. + duplicateFunction->replaceAllUsesWith(F); + duplicateFunction->eraseFromParent(); + + return extractedFunction; +} + +bool PartialInliner::runOnModule(Module& M) { + std::vector worklist; + worklist.reserve(M.size()); + for (Module::iterator FI = M.begin(), FE = M.end(); FI != FE; ++FI) + if (!FI->use_empty() && !FI->isDeclaration()) + worklist.push_back(&*FI); + + bool changed = false; + while (!worklist.empty()) { + Function* currFunc = worklist.back(); + worklist.pop_back(); + + if (currFunc->use_empty()) continue; + + bool recursive = false; + for (Function::use_iterator UI = currFunc->use_begin(), + UE = currFunc->use_end(); UI != UE; ++UI) + if (Instruction* I = dyn_cast(UI)) + if (I->getParent()->getParent() == currFunc) { + recursive = true; + break; + } + if (recursive) continue; + + + if (Function* newFunc = unswitchFunction(currFunc)) { + worklist.push_back(newFunc); + changed = true; + } + + } + + return changed; +} \ No newline at end of file diff --git a/lib/Transforms/Scalar/IndVarSimplify.cpp b/lib/Transforms/Scalar/IndVarSimplify.cpp index 83503fdea24..38b11985519 100644 --- a/lib/Transforms/Scalar/IndVarSimplify.cpp +++ b/lib/Transforms/Scalar/IndVarSimplify.cpp @@ -168,7 +168,7 @@ ICmpInst *IndVarSimplify::LinearFunctionTestReplace(Loop *L, // Expand the code for the iteration count into the preheader of the loop. BasicBlock *Preheader = L->getLoopPreheader(); - Value *ExitCnt = Rewriter.expandCodeFor(RHS, CmpIndVar->getType(), + Value *ExitCnt = Rewriter.expandCodeFor(RHS, IndVar->getType(), Preheader->getTerminator()); // Insert a new icmp_ne or icmp_eq instruction before the branch. @@ -392,10 +392,31 @@ bool IndVarSimplify::runOnLoop(Loop *L, LPPassManager &LPM) { // in this loop, insert a canonical induction variable of the largest size. Value *IndVar = 0; if (NeedCannIV) { + // Check to see if the loop already has a canonical-looking induction + // variable. If one is present and it's wider than the planned canonical + // induction variable, temporarily remove it, so that the Rewriter + // doesn't attempt to reuse it. + PHINode *OldCannIV = L->getCanonicalInductionVariable(); + if (OldCannIV) { + if (SE->getTypeSizeInBits(OldCannIV->getType()) > + SE->getTypeSizeInBits(LargestType)) + OldCannIV->removeFromParent(); + else + OldCannIV = 0; + } + IndVar = Rewriter.getOrInsertCanonicalInductionVariable(L,LargestType); + ++NumInserted; Changed = true; DOUT << "INDVARS: New CanIV: " << *IndVar; + + // Now that the official induction variable is established, reinsert + // the old canonical-looking variable after it so that the IR remains + // consistent. It will be deleted as part of the dead-PHI deletion at + // the end of the pass. + if (OldCannIV) + OldCannIV->insertAfter(cast(IndVar)); } // If we have a trip count expression, rewrite the loop's exit condition @@ -459,8 +480,8 @@ void IndVarSimplify::RewriteIVExpressions(Loop *L, const Type *LargestType, E = List.end(); UI != E; ++UI) { SCEVHandle Offset = UI->getOffset(); Value *Op = UI->getOperandValToReplace(); + const Type *UseTy = Op->getType(); Instruction *User = UI->getUser(); - bool isSigned = UI->isSigned(); // Compute the final addrec to expand into code. SCEVHandle AR = IU->getReplacementExpr(*UI); @@ -471,7 +492,7 @@ void IndVarSimplify::RewriteIVExpressions(Loop *L, const Type *LargestType, // Expand loop-invariant values in the loop preheader. They will // be sunk to the exit block later, if possible. NewVal = - Rewriter.expandCodeFor(AR, LargestType, + Rewriter.expandCodeFor(AR, UseTy, L->getLoopPreheader()->getTerminator()); Rewriter.setInsertionPoint(I); ++NumReplaced; @@ -485,74 +506,6 @@ void IndVarSimplify::RewriteIVExpressions(Loop *L, const Type *LargestType, if (!Stride->isLoopInvariant(L)) continue; - const Type *IVTy = Offset->getType(); - const Type *UseTy = Op->getType(); - - // Promote the Offset and Stride up to the canonical induction - // variable's bit width. - SCEVHandle PromotedOffset = Offset; - SCEVHandle PromotedStride = Stride; - if (SE->getTypeSizeInBits(IVTy) != SE->getTypeSizeInBits(LargestType)) { - // It doesn't matter for correctness whether zero or sign extension - // is used here, since the value is truncated away below, but if the - // value is signed, sign extension is more likely to be folded. - if (isSigned) { - PromotedOffset = SE->getSignExtendExpr(PromotedOffset, LargestType); - PromotedStride = SE->getSignExtendExpr(PromotedStride, LargestType); - } else { - PromotedOffset = SE->getZeroExtendExpr(PromotedOffset, LargestType); - // If the stride is obviously negative, use sign extension to - // produce things like x-1 instead of x+255. - if (isa(PromotedStride) && - cast(PromotedStride) - ->getValue()->getValue().isNegative()) - PromotedStride = SE->getSignExtendExpr(PromotedStride, - LargestType); - else - PromotedStride = SE->getZeroExtendExpr(PromotedStride, - LargestType); - } - } - - // Create the SCEV representing the offset from the canonical - // induction variable, still in the canonical induction variable's - // type, so that all expanded arithmetic is done in the same type. - SCEVHandle NewAR = SE->getAddRecExpr(SE->getIntegerSCEV(0, LargestType), - PromotedStride, L); - // Add the PromotedOffset as a separate step, because it may not be - // loop-invariant. - NewAR = SE->getAddExpr(NewAR, PromotedOffset); - - // Expand the addrec into instructions. - Value *V = Rewriter.expandCodeFor(NewAR); - - // Insert an explicit cast if necessary to truncate the value - // down to the original stride type. This is done outside of - // SCEVExpander because in SCEV expressions, a truncate of an - // addrec is always folded. - if (LargestType != IVTy) { - if (SE->getTypeSizeInBits(IVTy) != SE->getTypeSizeInBits(LargestType)) - NewAR = SE->getTruncateExpr(NewAR, IVTy); - if (Rewriter.isInsertedExpression(NewAR)) - V = Rewriter.expandCodeFor(NewAR); - else { - V = Rewriter.InsertCastOfTo(CastInst::getCastOpcode(V, false, - IVTy, false), - V, IVTy); - assert(!isa(V) && !isa(V) && - "LargestType wasn't actually the largest type!"); - // Force the rewriter to use this trunc whenever this addrec - // appears so that it doesn't insert new phi nodes or - // arithmetic in a different type. - Rewriter.addInsertedValue(V, NewAR); - } - } - - DOUT << "INDVARS: Made offset-and-trunc IV for offset " - << *IVTy << " " << *Offset << ": "; - DEBUG(WriteAsOperand(*DOUT, V, false)); - DOUT << "\n"; - // Now expand it into actual Instructions and patch it into place. NewVal = Rewriter.expandCodeFor(AR, UseTy); } diff --git a/lib/Transforms/Scalar/InstructionCombining.cpp b/lib/Transforms/Scalar/InstructionCombining.cpp index 6d2ff0e3e53..5465e4a8846 100644 --- a/lib/Transforms/Scalar/InstructionCombining.cpp +++ b/lib/Transforms/Scalar/InstructionCombining.cpp @@ -2608,21 +2608,6 @@ Instruction *InstCombiner::visitFSub(BinaryOperator &I) { else if (Op1I->getOperand(1) == Op0) // X-(Y+X) == -Y return BinaryOperator::CreateFNeg(Op1I->getOperand(0), I.getName()); } - - if (Op1I->hasOneUse()) { - // Replace (x - (y - z)) with (x + (z - y)) if the (y - z) subexpression - // is not used by anyone else... - // - if (Op1I->getOpcode() == Instruction::FSub) { - // Swap the two operands of the subexpr... - Value *IIOp0 = Op1I->getOperand(0), *IIOp1 = Op1I->getOperand(1); - Op1I->setOperand(0, IIOp1); - Op1I->setOperand(1, IIOp0); - - // Create the new top level fadd instruction... - return BinaryOperator::CreateFAdd(Op0, Op1); - } - } } return 0; @@ -11824,7 +11809,8 @@ Instruction *InstCombiner::visitStoreInst(StoreInst &SI) { if (SI.isVolatile()) return 0; // Don't hack volatile stores. // store X, null -> turns into 'unreachable' in SimplifyCFG - if (isa(Ptr)) { + if (isa(Ptr) && + cast(Ptr->getType())->getAddressSpace() == 0) { if (!isa(Val)) { SI.setOperand(0, UndefValue::get(Val->getType())); if (Instruction *U = dyn_cast(Val)) diff --git a/lib/Transforms/Scalar/SimplifyCFGPass.cpp b/lib/Transforms/Scalar/SimplifyCFGPass.cpp index b499279c696..5a85a047bec 100644 --- a/lib/Transforms/Scalar/SimplifyCFGPass.cpp +++ b/lib/Transforms/Scalar/SimplifyCFGPass.cpp @@ -125,13 +125,17 @@ static bool MarkAliveBlocks(BasicBlock *BB, } } - if (StoreInst *SI = dyn_cast(BBI)) - if (isa(SI->getOperand(1)) || - isa(SI->getOperand(1))) { + if (StoreInst *SI = dyn_cast(BBI)) { + Value *Ptr = SI->getOperand(1); + + if (isa(Ptr) || + (isa(Ptr) && + cast(Ptr->getType())->getAddressSpace() == 0)) { ChangeToUnreachable(SI); Changed = true; break; } + } } // Turn invokes that call 'nounwind' functions into ordinary calls. diff --git a/lib/VMCore/Function.cpp b/lib/VMCore/Function.cpp index 3a991f62d84..54bd895fd40 100644 --- a/lib/VMCore/Function.cpp +++ b/lib/VMCore/Function.cpp @@ -364,4 +364,15 @@ Function *Intrinsic::getDeclaration(Module *M, ID id, const Type **Tys, #include "llvm/Intrinsics.gen" #undef GET_LLVM_INTRINSIC_FOR_GCC_BUILTIN + /// hasAddressTaken - returns true if there are any uses of this function + /// other than direct calls or invokes to it. +bool Function::hasAddressTaken() const { + for (Value::use_const_iterator I = use_begin(), E = use_end(); I != E; ++I) { + if (I.getOperandNo() != 0 || + (!isa(*I) && !isa(*I))) + return true; + } + return false; +} + // vim: sw=2 ai diff --git a/lib/VMCore/Instruction.cpp b/lib/VMCore/Instruction.cpp index 7556b8ebe76..e0764e494b2 100644 --- a/lib/VMCore/Instruction.cpp +++ b/lib/VMCore/Instruction.cpp @@ -218,9 +218,12 @@ bool Instruction::isIdenticalTo(const Instruction *I) const { } // isSameOperationAs +// This should be kept in sync with isEquivalentOperation in +// lib/Transforms/IPO/MergeFunctions.cpp. bool Instruction::isSameOperationAs(const Instruction *I) const { - if (getOpcode() != I->getOpcode() || getType() != I->getType() || - getNumOperands() != I->getNumOperands()) + if (getOpcode() != I->getOpcode() || + getNumOperands() != I->getNumOperands() || + getType() != I->getType()) return false; // We have two instructions of identical opcode and #operands. Check to see diff --git a/lib/VMCore/Verifier.cpp b/lib/VMCore/Verifier.cpp index b1297ff2256..e9f2acda28d 100644 --- a/lib/VMCore/Verifier.cpp +++ b/lib/VMCore/Verifier.cpp @@ -276,8 +276,8 @@ namespace { int VT, unsigned ArgNo, std::string &Suffix); void VerifyIntrinsicPrototype(Intrinsic::ID ID, Function *F, unsigned RetNum, unsigned ParamNum, ...); - void VerifyAttrs(Attributes Attrs, const Type *Ty, - bool isReturnValue, const Value *V); + void VerifyParameterAttrs(Attributes Attrs, const Type *Ty, + bool isReturnValue, const Value *V); void VerifyFunctionAttrs(const FunctionType *FT, const AttrListPtr &Attrs, const Value *V); @@ -437,22 +437,23 @@ void Verifier::visitGlobalAlias(GlobalAlias &GA) { void Verifier::verifyTypeSymbolTable(TypeSymbolTable &ST) { } -// VerifyAttrs - Check the given parameter attributes for an argument or return +// VerifyParameterAttrs - Check the given attributes for an argument or return // value of the specified type. The value V is printed in error messages. -void Verifier::VerifyAttrs(Attributes Attrs, const Type *Ty, - bool isReturnValue, const Value *V) { +void Verifier::VerifyParameterAttrs(Attributes Attrs, const Type *Ty, + bool isReturnValue, const Value *V) { if (Attrs == Attribute::None) return; + Attributes FnCheckAttr = Attrs & Attribute::FunctionOnly; + Assert1(!FnCheckAttr, "Attribute " + Attribute::getAsString(FnCheckAttr) + + " only applies to the function!", V); + if (isReturnValue) { Attributes RetI = Attrs & Attribute::ParameterOnly; Assert1(!RetI, "Attribute " + Attribute::getAsString(RetI) + " does not apply to return values!", V); } - Attributes FnCheckAttr = Attrs & Attribute::FunctionOnly; - Assert1(!FnCheckAttr, "Attribute " + Attribute::getAsString(FnCheckAttr) + - " only applies to functions!", V); - + for (unsigned i = 0; i < array_lengthof(Attribute::MutuallyIncompatible); ++i) { Attributes MutI = Attrs & Attribute::MutuallyIncompatible[i]; @@ -495,9 +496,9 @@ void Verifier::VerifyFunctionAttrs(const FunctionType *FT, else if (Attr.Index-1 < FT->getNumParams()) Ty = FT->getParamType(Attr.Index-1); else - break; // VarArgs attributes, don't verify. - - VerifyAttrs(Attr.Attrs, Ty, Attr.Index == 0, V); + break; // VarArgs attributes, verified elsewhere. + + VerifyParameterAttrs(Attr.Attrs, Ty, Attr.Index == 0, V); if (Attr.Attrs & Attribute::Nest) { Assert1(!SawNest, "More than one parameter has attribute nest!", V); @@ -509,10 +510,10 @@ void Verifier::VerifyFunctionAttrs(const FunctionType *FT, } Attributes FAttrs = Attrs.getFnAttributes(); - Assert1(!(FAttrs & (~Attribute::FunctionOnly)), - "Attribute " + Attribute::getAsString(FAttrs) + - " does not apply to function!", V); - + Attributes NotFn = FAttrs & (~Attribute::FunctionOnly); + Assert1(!NotFn, "Attribute " + Attribute::getAsString(NotFn) + + " does not apply to the function!", V); + for (unsigned i = 0; i < array_lengthof(Attribute::MutuallyIncompatible); ++i) { Attributes MutI = FAttrs & Attribute::MutuallyIncompatible[i]; @@ -1025,7 +1026,7 @@ void Verifier::VerifyCallSite(CallSite CS) { for (unsigned Idx = 1 + FTy->getNumParams(); Idx <= CS.arg_size(); ++Idx) { Attributes Attr = Attrs.getParamAttributes(Idx); - VerifyAttrs(Attr, CS.getArgument(Idx-1)->getType(), false, I); + VerifyParameterAttrs(Attr, CS.getArgument(Idx-1)->getType(), false, I); Attributes VArgI = Attr & Attribute::VarArgsIncompatible; Assert1(!VArgI, "Attribute " + Attribute::getAsString(VArgI) + diff --git a/test/Bitcode/2009-06-11-FirstClassAggregateConstant.ll b/test/Bitcode/2009-06-11-FirstClassAggregateConstant.ll new file mode 100644 index 00000000000..415f88e1637 --- /dev/null +++ b/test/Bitcode/2009-06-11-FirstClassAggregateConstant.ll @@ -0,0 +1,12 @@ +; RUN: llvm-as < %s | llvm-dis -disable-output +; PR4373 + +@foo = weak global { i32 } zeroinitializer +@bar = weak global i32 0 + +define void @test() { +entry: + store { i32 } zeroinitializer, { i32 }* @foo + store i32 1, i32* @bar + ret void +} diff --git a/test/BugPoint/misopt-basictest.ll b/test/BugPoint/misopt-basictest.ll index 11cc2ae7995..462a637c2c2 100644 --- a/test/BugPoint/misopt-basictest.ll +++ b/test/BugPoint/misopt-basictest.ll @@ -1,4 +1,7 @@ ; RUN: bugpoint %s -dce -bugpoint-deletecalls -simplifycfg -silence-passes %bugpoint_topts +; XFAIL: powerpc-.*-linux +; END. +; Failure on PPC Linux is due to PR4293. @.LC0 = internal global [13 x i8] c"Hello World\0A\00" ; <[13 x i8]*> [#uses=1] diff --git a/test/CodeGen/ARM/2009-06-12-RegScavengerAssert.ll b/test/CodeGen/ARM/2009-06-12-RegScavengerAssert.ll new file mode 100644 index 00000000000..45b4bd48f51 --- /dev/null +++ b/test/CodeGen/ARM/2009-06-12-RegScavengerAssert.ll @@ -0,0 +1,77 @@ +; RUN: llvm-as < %s | llc -mtriple=armv6-apple-darwin + + type { i32, i32, %struct.D_Sym**, [3 x %struct.D_Sym*] } ; type %0 + type { i32, %struct.D_Reduction** } ; type %1 + type { i32, %struct.D_RightEpsilonHint* } ; type %2 + type { i32, %struct.D_ErrorRecoveryHint* } ; type %3 + type { i32, i32, %struct.D_Reduction**, [3 x %struct.D_Reduction*] } ; type %4 + %struct.D_ErrorRecoveryHint = type { i16, i16, i8* } + %struct.D_ParseNode = type { i32, %struct.d_loc_t, i8*, i8*, %struct.D_Scope*, void (%struct.D_Parser*, %struct.d_loc_t*, i8**)*, i8*, i8* } + %struct.D_Parser = type { i8*, void (%struct.D_Parser*, %struct.d_loc_t*, i8**)*, %struct.D_Scope*, void (%struct.D_Parser*)*, %struct.D_ParseNode* (%struct.D_Parser*, i32, %struct.D_ParseNode**)*, void (%struct.D_ParseNode*)*, %struct.d_loc_t, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } + %struct.D_ParserTables = type { i32, %struct.D_State*, i16*, i32, i32, %struct.D_Symbol*, void (%struct.D_Parser*, %struct.d_loc_t*, i8**)*, i32, %struct.D_Pass*, i32 } + %struct.D_Pass = type { i8*, i32, i32, i32 } + %struct.D_Reduction = type { i16, i16, i32 (i8*, i8**, i32, i32, %struct.D_Parser*)*, i32 (i8*, i8**, i32, i32, %struct.D_Parser*)*, i16, i16, i32, i32, i32, i32, i32 (i8*, i8**, i32, i32, %struct.D_Parser*)** } + %struct.D_RightEpsilonHint = type { i16, i16, %struct.D_Reduction* } + %struct.D_Scope = type { i8, %struct.D_Sym*, %struct.D_SymHash*, %struct.D_Sym*, %struct.D_Scope*, %struct.D_Scope*, %struct.D_Scope*, %struct.D_Scope*, %struct.D_Scope* } + %struct.D_Shift = type { i16, i8, i8, i32, i32, i32 (i8*, i8**, i32, i32, %struct.D_Parser*)* } + %struct.D_State = type { i8*, i32, %1, %2, %3, %struct.D_Shift**, i32 (i8**, i32*, i32*, i16*, i32*, i8*, i32*)*, i8*, i8, i8, i8, i8*, %struct.D_Shift***, i32 } + %struct.D_Sym = type { i8*, i32, i32, %struct.D_Sym*, %struct.D_Sym*, i32 } + %struct.D_SymHash = type { i32, i32, %0 } + %struct.D_Symbol = type { i32, i8*, i32 } + %struct.PNode = type { i32, i32, i32, i32, %struct.D_Reduction*, %struct.D_Shift*, i32, %struct.VecPNode, i32, i8, i8, %struct.PNode*, %struct.PNode*, %struct.PNode*, %struct.PNode*, i8*, i8*, %struct.D_Scope*, i8*, %struct.D_ParseNode } + %struct.PNodeHash = type { %struct.PNode**, i32, i32, i32, %struct.PNode* } + %struct.Parser = type { %struct.D_Parser, i8*, i8*, %struct.D_ParserTables*, i32, i32, i32, i32, i32, i32, i32, %struct.PNodeHash, %struct.SNodeHash, %struct.Reduction*, %struct.Shift*, %struct.D_Scope*, %struct.SNode*, i32, %struct.Reduction*, %struct.Shift*, i32, %struct.PNode*, %struct.SNode*, %struct.ZNode*, %4, %struct.ShiftResult*, %struct.D_Shift, %struct.Parser*, i8* } + %struct.Reduction = type { %struct.ZNode*, %struct.SNode*, %struct.D_Reduction*, %struct.SNode*, i32, %struct.Reduction* } + %struct.SNode = type { %struct.D_State*, %struct.D_Scope*, i8*, %struct.d_loc_t, i32, %struct.PNode*, %struct.VecZNode, i32, %struct.SNode*, %struct.SNode* } + %struct.SNodeHash = type { %struct.SNode**, i32, i32, i32, %struct.SNode*, %struct.SNode* } + %struct.Shift = type { %struct.SNode*, %struct.Shift* } + %struct.ShiftResult = type { %struct.D_Shift*, %struct.d_loc_t } + %struct.VecPNode = type { i32, i32, %struct.PNode**, [3 x %struct.PNode*] } + %struct.VecSNode = type { i32, i32, %struct.SNode**, [3 x %struct.SNode*] } + %struct.VecZNode = type { i32, i32, %struct.ZNode**, [3 x %struct.ZNode*] } + %struct.ZNode = type { %struct.PNode*, %struct.VecSNode } + %struct.d_loc_t = type { i8*, i8*, i32, i32, i32 } + +declare void @llvm.memcpy.i32(i8* nocapture, i8* nocapture, i32, i32) nounwind + +define fastcc i32 @exhaustive_parse(%struct.Parser* %p, i32 %state) nounwind { +entry: + store i8* undef, i8** undef, align 4 + %0 = getelementptr %struct.Parser* %p, i32 0, i32 0, i32 6 ; <%struct.d_loc_t*> [#uses=1] + %1 = bitcast %struct.d_loc_t* %0 to i8* ; [#uses=1] + call void @llvm.memcpy.i32(i8* undef, i8* %1, i32 20, i32 4) + br label %bb10 + +bb10: ; preds = %bb30, %bb29, %bb26, %entry + br i1 undef, label %bb18, label %bb20 + +bb18: ; preds = %bb10 + br i1 undef, label %bb20, label %bb19 + +bb19: ; preds = %bb18 + br label %bb20 + +bb20: ; preds = %bb19, %bb18, %bb10 + br i1 undef, label %bb21, label %bb22 + +bb21: ; preds = %bb20 + unreachable + +bb22: ; preds = %bb20 + br i1 undef, label %bb24, label %bb26 + +bb24: ; preds = %bb22 + unreachable + +bb26: ; preds = %bb22 + br i1 undef, label %bb10, label %bb29 + +bb29: ; preds = %bb26 + br i1 undef, label %bb10, label %bb30 + +bb30: ; preds = %bb29 + br i1 undef, label %bb31, label %bb10 + +bb31: ; preds = %bb30 + unreachable +} diff --git a/test/CodeGen/ARM/arguments_f64_backfill.ll b/test/CodeGen/ARM/arguments_f64_backfill.ll new file mode 100644 index 00000000000..07d928abe81 --- /dev/null +++ b/test/CodeGen/ARM/arguments_f64_backfill.ll @@ -0,0 +1,8 @@ +; RUN: llvm-as < %s | llc -mtriple=arm-linux-gnueabi -mattr=+vfp2 -float-abi=hard | grep {fcpys s0, s1} + +define float @f(float %z, double %a, float %b) { + %tmp = call float @g(float %b) + ret float %tmp +} + +declare float @g(float) diff --git a/test/CodeGen/ARM/lsr-code-insertion.ll b/test/CodeGen/ARM/lsr-code-insertion.ll index 0a9227982d4..3881e91453b 100644 --- a/test/CodeGen/ARM/lsr-code-insertion.ll +++ b/test/CodeGen/ARM/lsr-code-insertion.ll @@ -1,4 +1,4 @@ -; RUN: llvm-as < %s | llc -stats |& grep {40.*Number of machine instrs printed} +; RUN: llvm-as < %s | llc -stats |& grep {39.*Number of machine instrs printed} ; RUN: llvm-as < %s | llc -stats |& grep {.*Number of re-materialization} ; This test really wants to check that the resultant "cond_true" block only ; has a single store in it, and that cond_true55 only has code to materialize diff --git a/test/CodeGen/ARM/stm.ll b/test/CodeGen/ARM/stm.ll new file mode 100644 index 00000000000..585645b0214 --- /dev/null +++ b/test/CodeGen/ARM/stm.ll @@ -0,0 +1,13 @@ +; RUN: llvm-as < %s | llc -mtriple=arm-apple-darwin -mattr=+v6,+vfp2 -arm-pre-alloc-loadstore-opti | grep stm | count 2 + +@"\01LC" = internal constant [32 x i8] c"Boolean Not: %d %d %d %d %d %d\0A\00", section "__TEXT,__cstring,cstring_literals" ; <[32 x i8]*> [#uses=1] +@"\01LC1" = internal constant [26 x i8] c"Bitwise Not: %d %d %d %d\0A\00", section "__TEXT,__cstring,cstring_literals" ; <[26 x i8]*> [#uses=1] + +declare i32 @printf(i8* nocapture, ...) nounwind + +define i32 @main() nounwind { +entry: + %0 = tail call i32 (i8*, ...)* @printf(i8* getelementptr ([26 x i8]* @"\01LC1", i32 0, i32 0), i32 -2, i32 -3, i32 2, i32 -6) nounwind ; [#uses=0] + %1 = tail call i32 (i8*, ...)* @printf(i8* getelementptr ([32 x i8]* @"\01LC", i32 0, i32 0), i32 0, i32 1, i32 0, i32 1, i32 0, i32 1) nounwind ; [#uses=0] + ret i32 0 +} diff --git a/test/CodeGen/X86/2009-06-12-x86_64-tail-call-conv-out-of-sync-bug.ll b/test/CodeGen/X86/2009-06-12-x86_64-tail-call-conv-out-of-sync-bug.ll new file mode 100644 index 00000000000..001b7fc5a4a --- /dev/null +++ b/test/CodeGen/X86/2009-06-12-x86_64-tail-call-conv-out-of-sync-bug.ll @@ -0,0 +1,15 @@ +; RUN: llvm-as < %s | llc -tailcallopt -march=x86-64 -mattr=+sse2 -mtriple=x86_64-apple-darwin | grep fstpt +; RUN: llvm-as < %s | llc -tailcallopt -march=x86-64 -mattr=+sse2 -mtriple=x86_64-apple-darwin | grep xmm + +; Check that x86-64 tail calls support x86_fp80 and v2f32 types. (Tail call +; calling convention out of sync with standard c calling convention on x86_64) +; Bug 4278. + +declare fastcc double @tailcallee(x86_fp80, <2 x float>) + +define fastcc double @tailcall() { +entry: + %tmp = fpext float 1.000000e+00 to x86_fp80 + %tmp2 = tail call fastcc double @tailcallee( x86_fp80 %tmp, <2 x float> ) + ret double %tmp2 +} diff --git a/test/CodeGen/X86/tailcallstack64.ll b/test/CodeGen/X86/tailcallstack64.ll index d734065c005..c81327e5143 100644 --- a/test/CodeGen/X86/tailcallstack64.ll +++ b/test/CodeGen/X86/tailcallstack64.ll @@ -3,19 +3,19 @@ ; Move param %in1 to temp register (%eax). ; RUN: llvm-as < %s | llc -tailcallopt -march=x86-64 -x86-asm-syntax=att | grep {movl 40(%rsp), %eax} ; Add %in1 %p1 to another temporary register (%r9d). -; RUN: llvm-as < %s | llc -tailcallopt -march=x86-64 -x86-asm-syntax=att | grep {movl %edi, %r9d} -; RUN: llvm-as < %s | llc -tailcallopt -march=x86-64 -x86-asm-syntax=att | grep {addl 32(%rsp), %r9d} +; RUN: llvm-as < %s | llc -tailcallopt -march=x86-64 -x86-asm-syntax=att | grep {movl %edi, %r10d} +; RUN: llvm-as < %s | llc -tailcallopt -march=x86-64 -x86-asm-syntax=att | grep {addl 32(%rsp), %r10d} ; Move result of addition to stack. -; RUN: llvm-as < %s | llc -tailcallopt -march=x86-64 -x86-asm-syntax=att | grep {movl %r9d, 40(%rsp)} +; RUN: llvm-as < %s | llc -tailcallopt -march=x86-64 -x86-asm-syntax=att | grep {movl %r10d, 40(%rsp)} ; Move param %in2 to stack. ; RUN: llvm-as < %s | llc -tailcallopt -march=x86-64 -x86-asm-syntax=att | grep {movl %eax, 32(%rsp)} -declare fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %a, i32 %b) +declare fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %a, i32 %b) -define fastcc i32 @tailcaller(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %in1, i32 %in2) { +define fastcc i32 @tailcaller(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %in1, i32 %in2) { entry: %tmp = add i32 %in1, %p1 - %retval = tail call fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %in2,i32 %tmp) + %retval = tail call fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %in2,i32 %tmp) ret i32 %retval } diff --git a/test/DebugInfo/2009-06-12-Inline.ll b/test/DebugInfo/2009-06-12-Inline.ll new file mode 100644 index 00000000000..87c42d5a683 --- /dev/null +++ b/test/DebugInfo/2009-06-12-Inline.ll @@ -0,0 +1,94 @@ +; RUN: llvm-as < %s | llc -f -o /dev/null + %llvm.dbg.anchor.type = type { i32, i32 } + %llvm.dbg.basictype.type = type { i32, { }*, i8*, { }*, i32, i64, i64, i64, i32, i32 } + %llvm.dbg.compile_unit.type = type { i32, { }*, i32, i8*, i8*, i8*, i1, i1, i8*, i32 } + %llvm.dbg.composite.type = type { i32, { }*, i8*, { }*, i32, i64, i64, i64, i32, { }*, { }*, i32 } + %llvm.dbg.derivedtype.type = type { i32, { }*, i8*, { }*, i32, i64, i64, i64, i32, { }* } + %llvm.dbg.subprogram.type = type { i32, { }*, { }*, i8*, i8*, i8*, { }*, i32, { }*, i1, i1 } + %llvm.dbg.variable.type = type { i32, { }*, i8*, { }*, i32, { }* } + %struct._objc_cache = type opaque + %struct._objc_category = type { i8*, i8*, %struct._objc_method_list*, %struct._objc_method_list*, %struct._objc_protocol_list*, i32, %struct._prop_list_t* } + %struct._objc_class = type { %struct._objc_class*, %struct._objc_class*, i8*, i32, i32, i32, %struct._objc_ivar_list*, %struct._objc_method_list*, %struct._objc_cache*, %struct._objc_protocol_list*, i8*, %struct._objc_class_extension* } + %struct._objc_class_extension = type { i32, i8*, %struct._prop_list_t* } + %struct._objc_exception_data = type { [18 x i32], [4 x i8*] } + %struct._objc_ivar = type { i8*, i8*, i32 } + %struct._objc_ivar_list = type opaque + %struct._objc_method = type { %struct.objc_selector*, i8*, i8* } + %struct._objc_method_description = type { %struct.objc_selector*, i8* } + %struct._objc_method_description_list = type { i32, [0 x %struct._objc_method_description] } + %struct._objc_method_list = type opaque + %struct._objc_module = type { i32, i32, i8*, %struct._objc_symtab* } + %struct._objc_protocol = type { %struct._objc_protocol_extension*, i8*, %struct._objc_protocol_list*, %struct._objc_method_description_list*, %struct._objc_method_description_list* } + %struct._objc_protocol_extension = type { i32, %struct._objc_method_description_list*, %struct._objc_method_description_list*, %struct._prop_list_t* } + %struct._objc_protocol_list = type { %struct._objc_protocol_list*, i32, [0 x %struct._objc_protocol] } + %struct._objc_super = type <{ %struct.objc_object*, %struct.objc_class* }> + %struct._objc_symtab = type { i32, %struct.objc_selector*, i16, i16, [0 x i8*] } + %struct._prop_list_t = type { i32, i32, [0 x %struct._prop_t] } + %struct._prop_t = type { i8*, i8* } + %struct.objc_class = type opaque + %struct.objc_object = type opaque + %struct.objc_selector = type opaque +@"\01L_OBJC_IMAGE_INFO" = internal constant [2 x i32] [i32 0, i32 16], section "__OBJC, __image_info,regular" ; <[2 x i32]*> [#uses=1] +@llvm.dbg.compile_units = linkonce constant %llvm.dbg.anchor.type { i32 458752, i32 17 }, section "llvm.metadata" ; <%llvm.dbg.anchor.type*> [#uses=1] +@.str = internal constant [4 x i8] c"t.m\00", section "llvm.metadata" ; <[4 x i8]*> [#uses=1] +@.str1 = internal constant [20 x i8] c"/Volumes/work/Radar\00", section "llvm.metadata" ; <[20 x i8]*> [#uses=1] +@.str2 = internal constant [10 x i8] c"clang 1.0\00", section "llvm.metadata" ; <[10 x i8]*> [#uses=1] +@llvm.dbg.compile_unit = internal constant %llvm.dbg.compile_unit.type { i32 458769, { }* bitcast (%llvm.dbg.anchor.type* @llvm.dbg.compile_units to { }*), i32 16, i8* getelementptr ([4 x i8]* @.str, i32 0, i32 0), i8* getelementptr ([20 x i8]* @.str1, i32 0, i32 0), i8* getelementptr ([10 x i8]* @.str2, i32 0, i32 0), i1 true, i1 false, i8* null, i32 1 }, section "llvm.metadata" ; <%llvm.dbg.compile_unit.type*> [#uses=1] +@llvm.dbg.subprograms = linkonce constant %llvm.dbg.anchor.type { i32 458752, i32 46 }, section "llvm.metadata" ; <%llvm.dbg.anchor.type*> [#uses=1] +@.str3 = internal constant [3 x i8] c"f1\00", section "llvm.metadata" ; <[3 x i8]*> [#uses=1] +@llvm.dbg.subprogram = internal constant %llvm.dbg.subprogram.type { i32 458798, { }* bitcast (%llvm.dbg.anchor.type* @llvm.dbg.subprograms to { }*), { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i8* getelementptr ([3 x i8]* @.str3, i32 0, i32 0), i8* getelementptr ([3 x i8]* @.str3, i32 0, i32 0), i8* getelementptr ([3 x i8]* @.str3, i32 0, i32 0), { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i32 3, { }* null, i1 false, i1 true }, section "llvm.metadata" ; <%llvm.dbg.subprogram.type*> [#uses=1] +@.str4 = internal constant [4 x i8] c"int\00", section "llvm.metadata" ; <[4 x i8]*> [#uses=1] +@llvm.dbg.basictype = internal constant %llvm.dbg.basictype.type { i32 458788, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i8* getelementptr ([4 x i8]* @.str4, i32 0, i32 0), { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i32 0, i64 32, i64 32, i64 0, i32 0, i32 5 }, section "llvm.metadata" ; <%llvm.dbg.basictype.type*> [#uses=1] +@llvm.dbg.array = internal constant [2 x { }*] [{ }* null, { }* bitcast (%llvm.dbg.basictype.type* @llvm.dbg.basictype to { }*)], section "llvm.metadata" ; <[2 x { }*]*> [#uses=1] +@llvm.dbg.composite = internal constant %llvm.dbg.composite.type { i32 458773, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i8* null, { }* null, i32 0, i64 0, i64 0, i64 0, i32 0, { }* null, { }* bitcast ([2 x { }*]* @llvm.dbg.array to { }*), i32 0 }, section "llvm.metadata" ; <%llvm.dbg.composite.type*> [#uses=1] +@llvm.dbg.derivedtype = internal constant %llvm.dbg.derivedtype.type { i32 458767, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i8* null, { }* null, i32 0, i64 32, i64 32, i64 0, i32 0, { }* bitcast (%llvm.dbg.composite.type* @llvm.dbg.composite to { }*) }, section "llvm.metadata" ; <%llvm.dbg.derivedtype.type*> [#uses=1] +@.str5 = internal constant [3 x i8] c"l0\00", section "llvm.metadata" ; <[3 x i8]*> [#uses=1] +@llvm.dbg.variable = internal constant %llvm.dbg.variable.type { i32 459008, { }* bitcast (%llvm.dbg.subprogram.type* @llvm.dbg.subprogram to { }*), i8* getelementptr ([3 x i8]* @.str5, i32 0, i32 0), { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i32 5, { }* bitcast (%llvm.dbg.derivedtype.type* @llvm.dbg.derivedtype to { }*) }, section "llvm.metadata" ; <%llvm.dbg.variable.type*> [#uses=1] +@.str6 = internal constant [3 x i8] c"f0\00", section "llvm.metadata" ; <[3 x i8]*> [#uses=1] +@llvm.dbg.subprogram7 = internal constant %llvm.dbg.subprogram.type { i32 458798, { }* bitcast (%llvm.dbg.anchor.type* @llvm.dbg.subprograms to { }*), { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i8* getelementptr ([3 x i8]* @.str6, i32 0, i32 0), i8* getelementptr ([3 x i8]* @.str6, i32 0, i32 0), i8* getelementptr ([3 x i8]* @.str6, i32 0, i32 0), { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i32 2, { }* null, i1 false, i1 true }, section "llvm.metadata" ; <%llvm.dbg.subprogram.type*> [#uses=1] +@.str8 = internal constant [2 x i8] c"x\00", section "llvm.metadata" ; <[2 x i8]*> [#uses=1] +@llvm.dbg.variable9 = internal constant %llvm.dbg.variable.type { i32 459009, { }* bitcast (%llvm.dbg.subprogram.type* @llvm.dbg.subprogram7 to { }*), i8* getelementptr ([2 x i8]* @.str8, i32 0, i32 0), { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i32 2, { }* bitcast (%llvm.dbg.basictype.type* @llvm.dbg.basictype to { }*) }, section "llvm.metadata" ; <%llvm.dbg.variable.type*> [#uses=1] +@"\01L_OBJC_CLASS_NAME_" = internal global [1 x i8] zeroinitializer, section "__TEXT,__cstring,cstring_literals", align 1 ; <[1 x i8]*> [#uses=1] +@"\01L_OBJC_MODULES" = internal global %struct._objc_module { i32 7, i32 16, i8* getelementptr ([1 x i8]* @"\01L_OBJC_CLASS_NAME_", i32 0, i32 0), %struct._objc_symtab* null }, section "__OBJC,__module_info,regular,no_dead_strip", align 4 ; <%struct._objc_module*> [#uses=1] +@llvm.used = appending global [3 x i8*] [i8* bitcast ([2 x i32]* @"\01L_OBJC_IMAGE_INFO" to i8*), i8* getelementptr ([1 x i8]* @"\01L_OBJC_CLASS_NAME_", i32 0, i32 0), i8* bitcast (%struct._objc_module* @"\01L_OBJC_MODULES" to i8*)], section "llvm.metadata" ; <[3 x i8*]*> [#uses=0] + +define void @f1() nounwind { +entry: + %x.addr.i = alloca i32 ; [#uses=2] + %l0 = alloca void (i32)*, align 4 ; [#uses=2] + call void @llvm.dbg.func.start({ }* bitcast (%llvm.dbg.subprogram.type* @llvm.dbg.subprogram to { }*)) + call void @llvm.dbg.stoppoint(i32 4, i32 3, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*)) + call void @llvm.dbg.func.start({ }* bitcast (%llvm.dbg.subprogram.type* @llvm.dbg.subprogram7 to { }*)) + store i32 1, i32* %x.addr.i + %0 = bitcast i32* %x.addr.i to { }* ; <{ }*> [#uses=1] + call void @llvm.dbg.declare({ }* %0, { }* bitcast (%llvm.dbg.variable.type* @llvm.dbg.variable9 to { }*)) + call void @llvm.dbg.stoppoint(i32 2, i32 66, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*)) + call void @llvm.dbg.stoppoint(i32 5, i32 3, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*)) + call void @llvm.dbg.region.end({ }* bitcast (%llvm.dbg.subprogram.type* @llvm.dbg.subprogram7 to { }*)) + %1 = bitcast void (i32)** %l0 to { }* ; <{ }*> [#uses=1] + call void @llvm.dbg.declare({ }* %1, { }* bitcast (%llvm.dbg.variable.type* @llvm.dbg.variable to { }*)) + store void (i32)* @f0, void (i32)** %l0 + call void @llvm.dbg.stoppoint(i32 6, i32 1, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*)) + call void @llvm.dbg.region.end({ }* bitcast (%llvm.dbg.subprogram.type* @llvm.dbg.subprogram to { }*)) + ret void +} + +declare void @llvm.dbg.func.start({ }*) nounwind readnone + +declare void @llvm.dbg.stoppoint(i32, i32, { }*) nounwind readnone + +define internal void @f0(i32 %x) nounwind alwaysinline { +entry: + %x.addr = alloca i32 ; [#uses=2] + call void @llvm.dbg.func.start({ }* bitcast (%llvm.dbg.subprogram.type* @llvm.dbg.subprogram7 to { }*)) + store i32 %x, i32* %x.addr + %0 = bitcast i32* %x.addr to { }* ; <{ }*> [#uses=1] + call void @llvm.dbg.declare({ }* %0, { }* bitcast (%llvm.dbg.variable.type* @llvm.dbg.variable9 to { }*)) + call void @llvm.dbg.stoppoint(i32 2, i32 66, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*)) + call void @llvm.dbg.region.end({ }* bitcast (%llvm.dbg.subprogram.type* @llvm.dbg.subprogram7 to { }*)) + ret void +} + +declare void @llvm.dbg.declare({ }*, { }*) nounwind readnone + +declare void @llvm.dbg.region.end({ }*) nounwind readnone diff --git a/test/DebugInfo/2009-06-12-InlineFuncStart.ll b/test/DebugInfo/2009-06-12-InlineFuncStart.ll new file mode 100644 index 00000000000..32e20ff8c91 --- /dev/null +++ b/test/DebugInfo/2009-06-12-InlineFuncStart.ll @@ -0,0 +1,75 @@ +; RUN: llvm-as < %s | llc +; RUN: llvm-as < %s | llc -O0 + %llvm.dbg.anchor.type = type { i32, i32 } + %llvm.dbg.basictype.type = type { i32, { }*, i8*, { }*, i32, i64, i64, i64, i32, i32 } + %llvm.dbg.compile_unit.type = type { i32, { }*, i32, i8*, i8*, i8* } + %llvm.dbg.subprogram.type = type { i32, { }*, { }*, i8*, i8*, i8*, { }*, i32, { }*, i1, i1 } +@llvm.dbg.subprogram = internal constant %llvm.dbg.subprogram.type { i32 393262, { }* bitcast (%llvm.dbg.anchor.type* @llvm.dbg.subprograms to { }*), { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i8* getelementptr ([4 x i8]* @.str3, i32 0, i32 0), i8* getelementptr ([4 x i8]* @.str3, i32 0, i32 0), i8* null, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i32 1, { }* bitcast (%llvm.dbg.basictype.type* @llvm.dbg.basictype to { }*), i1 false, i1 true }, section "llvm.metadata" ; <%llvm.dbg.subprogram.type*> [#uses=1] +@llvm.dbg.subprograms = linkonce constant %llvm.dbg.anchor.type { i32 393216, i32 46 }, section "llvm.metadata" ; <%llvm.dbg.anchor.type*> [#uses=1] +@llvm.dbg.compile_unit = internal constant %llvm.dbg.compile_unit.type { i32 393233, { }* bitcast (%llvm.dbg.anchor.type* @llvm.dbg.compile_units to { }*), i32 1, i8* getelementptr ([4 x i8]* @.str, i32 0, i32 0), i8* getelementptr ([5 x i8]* @.str1, i32 0, i32 0), i8* getelementptr ([52 x i8]* @.str2, i32 0, i32 0) }, section "llvm.metadata" ; <%llvm.dbg.compile_unit.type*> [#uses=1] +@llvm.dbg.compile_units = linkonce constant %llvm.dbg.anchor.type { i32 393216, i32 17 }, section "llvm.metadata" ; <%llvm.dbg.anchor.type*> [#uses=1] +@.str = internal constant [4 x i8] c"a.c\00", section "llvm.metadata" ; <[4 x i8]*> [#uses=1] +@.str1 = internal constant [5 x i8] c"/tmp\00", section "llvm.metadata" ; <[5 x i8]*> [#uses=1] +@.str2 = internal constant [52 x i8] c"4.2.1 (Based on Apple Inc. build 5627) (LLVM build)\00", section "llvm.metadata" ; <[52 x i8]*> [#uses=1] +@.str3 = internal constant [4 x i8] c"foo\00", section "llvm.metadata" ; <[4 x i8]*> [#uses=1] +@llvm.dbg.basictype = internal constant %llvm.dbg.basictype.type { i32 393252, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i8* getelementptr ([4 x i8]* @.str4, i32 0, i32 0), { }* null, i32 0, i64 32, i64 32, i64 0, i32 0, i32 5 }, section "llvm.metadata" ; <%llvm.dbg.basictype.type*> [#uses=1] +@.str4 = internal constant [4 x i8] c"int\00", section "llvm.metadata" ; <[4 x i8]*> [#uses=1] +@llvm.dbg.subprogram5 = internal constant %llvm.dbg.subprogram.type { i32 393262, { }* bitcast (%llvm.dbg.anchor.type* @llvm.dbg.subprograms to { }*), { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i8* getelementptr ([5 x i8]* @.str6, i32 0, i32 0), i8* getelementptr ([5 x i8]* @.str6, i32 0, i32 0), i8* null, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*), i32 2, { }* bitcast (%llvm.dbg.basictype.type* @llvm.dbg.basictype to { }*), i1 false, i1 true }, section "llvm.metadata" ; <%llvm.dbg.subprogram.type*> [#uses=1] +@.str6 = internal constant [5 x i8] c"main\00", section "llvm.metadata" ; <[5 x i8]*> [#uses=1] + +define i32 @foo() nounwind alwaysinline { +entry: + %retval = alloca i32 ; [#uses=2] + %0 = alloca i32 ; [#uses=2] + %"alloca point" = bitcast i32 0 to i32 ; [#uses=0] + call void @llvm.dbg.func.start({ }* bitcast (%llvm.dbg.subprogram.type* @llvm.dbg.subprogram to { }*)) + call void @llvm.dbg.stoppoint(i32 1, i32 0, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*)) + store i32 42, i32* %0, align 4 + %1 = load i32* %0, align 4 ; [#uses=1] + store i32 %1, i32* %retval, align 4 + br label %return + +return: ; preds = %entry + %retval1 = load i32* %retval ; [#uses=1] + call void @llvm.dbg.stoppoint(i32 1, i32 0, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*)) + call void @llvm.dbg.region.end({ }* bitcast (%llvm.dbg.subprogram.type* @llvm.dbg.subprogram to { }*)) + ret i32 %retval1 +} + +declare void @llvm.dbg.func.start({ }*) nounwind + +declare void @llvm.dbg.stoppoint(i32, i32, { }*) nounwind + +declare void @llvm.dbg.region.end({ }*) nounwind + +define i32 @main() nounwind { +entry: + %retval.i = alloca i32 ; [#uses=2] + %0 = alloca i32 ; [#uses=2] + %retval = alloca i32 ; [#uses=2] + %1 = alloca i32 ; [#uses=2] + %"alloca point" = bitcast i32 0 to i32 ; [#uses=0] + call void @llvm.dbg.func.start({ }* bitcast (%llvm.dbg.subprogram.type* @llvm.dbg.subprogram5 to { }*)) + br label %bb1 + +return: ; preds = %entry + %retval1 = load i32* %retval ; [#uses=1] + call void @llvm.dbg.stoppoint(i32 2, i32 0, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*)) + call void @llvm.dbg.region.end({ }* bitcast (%llvm.dbg.subprogram.type* @llvm.dbg.subprogram5 to { }*)) + ret i32 %retval1 + +bb1: + call void @llvm.dbg.stoppoint(i32 2, i32 0, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*)) + call void @llvm.dbg.func.start({ }* bitcast (%llvm.dbg.subprogram.type* @llvm.dbg.subprogram to { }*)) nounwind + call void @llvm.dbg.stoppoint(i32 1, i32 0, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*)) nounwind + store i32 42, i32* %0, align 4 + %2 = load i32* %0, align 4 ; [#uses=1] + store i32 %2, i32* %retval.i, align 4 + %retval1.i = load i32* %retval.i ; [#uses=1] + call void @llvm.dbg.stoppoint(i32 1, i32 0, { }* bitcast (%llvm.dbg.compile_unit.type* @llvm.dbg.compile_unit to { }*)) nounwind + call void @llvm.dbg.region.end({ }* bitcast (%llvm.dbg.subprogram.type* @llvm.dbg.subprogram to { }*)) nounwind + store i32 %retval1.i, i32* %1, align 4 + %3 = load i32* %1, align 4 ; [#uses=1] + store i32 %3, i32* %retval, align 4 + br label %return +} diff --git a/test/FrontendC/pr3518.c b/test/FrontendC/pr3518.c new file mode 100644 index 00000000000..4c193c7739b --- /dev/null +++ b/test/FrontendC/pr3518.c @@ -0,0 +1,24 @@ +// RUN: %llvmgcc %s -S -emit-llvm -O0 -o - | grep {= internal global} | count 4 +// PR 3518 +// Some of the objects were coming out as unintialized (external) before 3518 +// was fixed. Internal names are different between llvm-gcc and clang so they +// are not tested. + +extern void abort (void); + +struct A { int i; int j; }; +struct B { struct A *a; struct A *b; }; +struct C { struct B *c; struct A *d; }; +struct C e = { &(struct B) { &(struct A) { 1, 2 }, &(struct A) { 3, 4 } }, &(struct A) { 5, 6 } }; + +int +main (void) +{ + if (e.c->a->i != 1 || e.c->a->j != 2) + abort (); + if (e.c->b->i != 3 || e.c->b->j != 4) + abort (); + if (e.d->i != 5 || e.d->j != 6) + abort (); + return 0; +} diff --git a/test/FrontendC/pr4349.c b/test/FrontendC/pr4349.c new file mode 100644 index 00000000000..890482ea06a --- /dev/null +++ b/test/FrontendC/pr4349.c @@ -0,0 +1,39 @@ +// RUN: %llvmgcc %s -S -emit-llvm -O0 -o - | grep svars2 | grep {\\\[2 x \\\[2 x i8\\\]\\\]} +// RUN: %llvmgcc %s -S -emit-llvm -O0 -o - | grep svars2 | grep {, i\[\[:digit:\]\]\\+ 1)} | count 1 +// RUN: %llvmgcc %s -S -emit-llvm -O0 -o - | grep svars3 | grep {\\\[2 x i16\\\]} +// RUN: %llvmgcc %s -S -emit-llvm -O0 -o - | grep svars3 | grep {, i\[\[:digit:\]\]\\+ 1)} | count 1 +// RUN: %llvmgcc %s -S -emit-llvm -O0 -o - | grep svars4 | grep {\\\[2 x \\\[2 x i8\\\]\\\]} | count 1 +// RUN: %llvmgcc %s -S -emit-llvm -O0 -o - | grep svars4 | grep {, i\[\[:digit:\]\]\\+ 1, i\[\[:digit:\]\]\\+ 1)} | count 1 +// PR 4349 + +union reg +{ + unsigned char b[2][2]; + unsigned short w[2]; + unsigned int d; +}; +struct cpu +{ + union reg pc; +}; +extern struct cpu cpu; +struct svar +{ + void *ptr; +}; +struct svar svars1[] = +{ + { &((cpu.pc).w[0]) } +}; +struct svar svars2[] = +{ + { &((cpu.pc).b[0][1]) } +}; +struct svar svars3[] = +{ + { &((cpu.pc).w[1]) } +}; +struct svar svars4[] = +{ + { &((cpu.pc).b[1][1]) } +}; diff --git a/test/TableGen/Slice.td b/test/TableGen/Slice.td new file mode 100644 index 00000000000..cd9c6da1539 --- /dev/null +++ b/test/TableGen/Slice.td @@ -0,0 +1,87 @@ +// RUN: tblgen %s | grep {\\\[(set} | count 2 +// RUN: tblgen %s | grep {\\\[\\\]} | count 2 + +class ValueType { + int Size = size; + int Value = value; +} + +def f32 : ValueType<32, 1>; // 2 x i64 vector value + +class Intrinsic { + string Name = name; +} + +class Inst opcode, dag oopnds, dag iopnds, string asmstr, + list pattern> { + bits<8> Opcode = opcode; + dag OutOperands = oopnds; + dag InOperands = iopnds; + string AssemblyString = asmstr; + list Pattern = pattern; +} + +def ops; +def outs; +def ins; + +def set; + +// Define registers +class Register { + string Name = n; +} + +class RegisterClass regTypes, list regList> { + list RegTypes = regTypes; + list MemberList = regList; +} + +def XMM0: Register<"xmm0">; +def XMM1: Register<"xmm1">; +def XMM2: Register<"xmm2">; +def XMM3: Register<"xmm3">; +def XMM4: Register<"xmm4">; +def XMM5: Register<"xmm5">; +def XMM6: Register<"xmm6">; +def XMM7: Register<"xmm7">; +def XMM8: Register<"xmm8">; +def XMM9: Register<"xmm9">; +def XMM10: Register<"xmm10">; +def XMM11: Register<"xmm11">; +def XMM12: Register<"xmm12">; +def XMM13: Register<"xmm13">; +def XMM14: Register<"xmm14">; +def XMM15: Register<"xmm15">; + +def FR32 : RegisterClass<[f32], + [XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7, + XMM8, XMM9, XMM10, XMM11, + XMM12, XMM13, XMM14, XMM15]>; + +class SDNode {} +def not : SDNode; + +multiclass scalar opcode, string asmstr = "", list> patterns = []> { + def SSrr : Inst,patterns[0])>; + def SSrm : Inst,!if(!null(!cdr(patterns)),patterns[0],patterns[1]))>; +} + +multiclass vscalar opcode, string asmstr = "", list> patterns = []> { + def V#NAME#SSrr : Inst,patterns[0])>; + def V#NAME#SSrm : Inst,!if(!null(!cdr(patterns)),patterns[0],patterns[1]))>; +} + +multiclass myscalar opcode, string asmstr = "", list> patterns = []> : + scalar, + vscalar; + +defm NOT : myscalar<0x10, "not", [[], [(set FR32:$dst, (f32 (not FR32:$src)))]]>; diff --git a/test/TableGen/if.td b/test/TableGen/if.td index 3c45d95eb20..9b2438245db 100644 --- a/test/TableGen/if.td +++ b/test/TableGen/if.td @@ -1,5 +1,5 @@ -// RUN: tblgen %s | grep {1, 2, 3} | count 4 -// RUN: tblgen %s | grep {4, 5, 6} | count 2 +// RUN: tblgen %s | grep {\\\[1, 2, 3\\\]} | count 4 +// RUN: tblgen %s | grep {\\\[4, 5, 6\\\]} | count 2 class A> vals> { list first = vals[0]; diff --git a/test/Transforms/IndVarSimplify/2003-09-23-NotAtTop.ll b/test/Transforms/IndVarSimplify/2003-09-23-NotAtTop.ll index d70e6049797..b4a2c506189 100644 --- a/test/Transforms/IndVarSimplify/2003-09-23-NotAtTop.ll +++ b/test/Transforms/IndVarSimplify/2003-09-23-NotAtTop.ll @@ -1,4 +1,4 @@ -; RUN: llvm-as < %s | opt -indvars | llvm-dis | %prcontext Loop: 1 | grep %indvar +; RUN: llvm-as < %s | opt -indvars | llvm-dis | %prcontext ^Loop: 1 | grep %Canonical ; The indvar simplification code should ensure that the first PHI in the block ; is the canonical one! diff --git a/test/Transforms/IndVarSimplify/masked-iv.ll b/test/Transforms/IndVarSimplify/masked-iv.ll new file mode 100644 index 00000000000..c7583c9655a --- /dev/null +++ b/test/Transforms/IndVarSimplify/masked-iv.ll @@ -0,0 +1,24 @@ +; RUN: llvm-as < %s | opt -indvars | llvm-dis | grep trunc | count 1 + +; Indvars should do the IV arithmetic in the canonical IV type (i64), +; and only use one truncation. + +define void @foo(i64* %A, i64* %B, i64 %n, i64 %a, i64 %s) nounwind { +entry: + %t0 = icmp sgt i64 %n, 0 ; [#uses=1] + br i1 %t0, label %bb.preheader, label %return + +bb.preheader: ; preds = %entry + br label %bb + +bb: ; preds = %bb, %bb.preheader + %i.01 = phi i64 [ %t6, %bb ], [ %a, %bb.preheader ] ; [#uses=3] + %t1 = and i64 %i.01, 255 ; [#uses=1] + %t2 = getelementptr i64* %A, i64 %t1 ; [#uses=1] + store i64 %i.01, i64* %t2, align 8 + %t6 = add i64 %i.01, %s ; [#uses=1] + br label %bb + +return: ; preds = %entry + ret void +} diff --git a/test/Transforms/InstCombine/2009-06-11-StoreAddrSpace.ll b/test/Transforms/InstCombine/2009-06-11-StoreAddrSpace.ll new file mode 100644 index 00000000000..e86fed3c364 --- /dev/null +++ b/test/Transforms/InstCombine/2009-06-11-StoreAddrSpace.ll @@ -0,0 +1,7 @@ +; RUN: llvm-as < %s | opt -instcombine | llvm-dis | grep store +; PR4366 + +define void @a() { + store i32 0, i32 addrspace(1)* null + ret void +} diff --git a/test/Transforms/InstCombine/fsub-fsub.ll b/test/Transforms/InstCombine/fsub-fsub.ll new file mode 100644 index 00000000000..ab70479204b --- /dev/null +++ b/test/Transforms/InstCombine/fsub-fsub.ll @@ -0,0 +1,8 @@ +; RUN: llvm-as < %s | opt -instcombine | llvm-dis | grep fsub | count 2 +; PR4374 + +define float @func(float %a, float %b) nounwind { + %tmp3 = fsub float %a, %b + %tmp4 = fsub float -0.000000e+00, %tmp3 + ret float %tmp4 +} diff --git a/test/Transforms/MergeFunc/fold-weak.ll b/test/Transforms/MergeFunc/fold-weak.ll index d98fde0228f..cea49fb1cec 100644 --- a/test/Transforms/MergeFunc/fold-weak.ll +++ b/test/Transforms/MergeFunc/fold-weak.ll @@ -1,4 +1,6 @@ -; RUN: llvm-as < %s | opt -mergefunc | llvm-dis | grep {alias weak} | count 2 +; RUN: llvm-as < %s | opt -mergefunc | llvm-dis > %t +; RUN: grep {define weak} %t | count 2 +; RUN: grep {call} %t | count 2 define weak i32 @sum(i32 %x, i32 %y) { %sum = add i32 %x, %y diff --git a/test/lib/llvm.exp b/test/lib/llvm.exp index d2ddb8f757f..3e2632f12c1 100644 --- a/test/lib/llvm.exp +++ b/test/lib/llvm.exp @@ -184,7 +184,7 @@ proc RunLLVMTests { test_source_files } { #split up target if more then 1 specified foreach target [split $targets ,] { - if { [regexp {\*} $target match] } { + if { $target == "*" } { if {$targetPASS != 1} { set outcome XFAIL } diff --git a/tools/lto/LTOCodeGenerator.cpp b/tools/lto/LTOCodeGenerator.cpp index 0bd2abe2455..6f8a02867c3 100644 --- a/tools/lto/LTOCodeGenerator.cpp +++ b/tools/lto/LTOCodeGenerator.cpp @@ -248,7 +248,7 @@ bool LTOCodeGenerator::assemble(const std::string& asmPath, std::vector args; std::string targetTriple = _linker.getModule()->getTargetTriple(); args.push_back(tool.c_str()); - if ( targetTriple.find("darwin") != targetTriple.size() ) { + if ( targetTriple.find("darwin") != std::string::npos ) { // darwin specific command line options if (strncmp(targetTriple.c_str(), "i386-apple-", 11) == 0) { args.push_back("-arch"); diff --git a/unittests/ADT/TripleTest.cpp b/unittests/ADT/TripleTest.cpp index 771d1c75d91..dee09f46ee9 100644 --- a/unittests/ADT/TripleTest.cpp +++ b/unittests/ADT/TripleTest.cpp @@ -18,120 +18,120 @@ TEST(TripleTest, BasicParsing) { Triple T; T = Triple(""); - EXPECT_EQ(T.getArchName(), ""); - EXPECT_EQ(T.getVendorName(), ""); - EXPECT_EQ(T.getOSName(), ""); - EXPECT_EQ(T.getEnvironmentName(), ""); + EXPECT_EQ("", T.getArchName()); + EXPECT_EQ("", T.getVendorName()); + EXPECT_EQ("", T.getOSName()); + EXPECT_EQ("", T.getEnvironmentName()); T = Triple("-"); - EXPECT_EQ(T.getArchName(), ""); - EXPECT_EQ(T.getVendorName(), ""); - EXPECT_EQ(T.getOSName(), ""); - EXPECT_EQ(T.getEnvironmentName(), ""); + EXPECT_EQ("", T.getArchName()); + EXPECT_EQ("", T.getVendorName()); + EXPECT_EQ("", T.getOSName()); + EXPECT_EQ("", T.getEnvironmentName()); T = Triple("--"); - EXPECT_EQ(T.getArchName(), ""); - EXPECT_EQ(T.getVendorName(), ""); - EXPECT_EQ(T.getOSName(), ""); - EXPECT_EQ(T.getEnvironmentName(), ""); + EXPECT_EQ("", T.getArchName()); + EXPECT_EQ("", T.getVendorName()); + EXPECT_EQ("", T.getOSName()); + EXPECT_EQ("", T.getEnvironmentName()); T = Triple("---"); - EXPECT_EQ(T.getArchName(), ""); - EXPECT_EQ(T.getVendorName(), ""); - EXPECT_EQ(T.getOSName(), ""); - EXPECT_EQ(T.getEnvironmentName(), ""); + EXPECT_EQ("", T.getArchName()); + EXPECT_EQ("", T.getVendorName()); + EXPECT_EQ("", T.getOSName()); + EXPECT_EQ("", T.getEnvironmentName()); T = Triple("----"); - EXPECT_EQ(T.getArchName(), ""); - EXPECT_EQ(T.getVendorName(), ""); - EXPECT_EQ(T.getOSName(), ""); - EXPECT_EQ(T.getEnvironmentName(), "-"); + EXPECT_EQ("", T.getArchName()); + EXPECT_EQ("", T.getVendorName()); + EXPECT_EQ("", T.getOSName()); + EXPECT_EQ("-", T.getEnvironmentName()); T = Triple("a"); - EXPECT_EQ(T.getArchName(), "a"); - EXPECT_EQ(T.getVendorName(), ""); - EXPECT_EQ(T.getOSName(), ""); - EXPECT_EQ(T.getEnvironmentName(), ""); + EXPECT_EQ("a", T.getArchName()); + EXPECT_EQ("", T.getVendorName()); + EXPECT_EQ("", T.getOSName()); + EXPECT_EQ("", T.getEnvironmentName()); T = Triple("a-b"); - EXPECT_EQ(T.getArchName(), "a"); - EXPECT_EQ(T.getVendorName(), "b"); - EXPECT_EQ(T.getOSName(), ""); - EXPECT_EQ(T.getEnvironmentName(), ""); + EXPECT_EQ("a", T.getArchName()); + EXPECT_EQ("b", T.getVendorName()); + EXPECT_EQ("", T.getOSName()); + EXPECT_EQ("", T.getEnvironmentName()); T = Triple("a-b-c"); - EXPECT_EQ(T.getArchName(), "a"); - EXPECT_EQ(T.getVendorName(), "b"); - EXPECT_EQ(T.getOSName(), "c"); - EXPECT_EQ(T.getEnvironmentName(), ""); + EXPECT_EQ("a", T.getArchName()); + EXPECT_EQ("b", T.getVendorName()); + EXPECT_EQ("c", T.getOSName()); + EXPECT_EQ("", T.getEnvironmentName()); T = Triple("a-b-c-d"); - EXPECT_EQ(T.getArchName(), "a"); - EXPECT_EQ(T.getVendorName(), "b"); - EXPECT_EQ(T.getOSName(), "c"); - EXPECT_EQ(T.getEnvironmentName(), "d"); + EXPECT_EQ("a", T.getArchName()); + EXPECT_EQ("b", T.getVendorName()); + EXPECT_EQ("c", T.getOSName()); + EXPECT_EQ("d", T.getEnvironmentName()); } TEST(TripleTest, ParsedIDs) { Triple T; T = Triple("i386-apple-darwin"); - EXPECT_EQ(T.getArch(), Triple::x86); - EXPECT_EQ(T.getVendor(), Triple::Apple); - EXPECT_EQ(T.getOS(), Triple::Darwin); + EXPECT_EQ(Triple::x86, T.getArch()); + EXPECT_EQ(Triple::Apple, T.getVendor()); + EXPECT_EQ(Triple::Darwin, T.getOS()); T = Triple("x86_64-pc-linux-gnu"); - EXPECT_EQ(T.getArch(), Triple::x86_64); - EXPECT_EQ(T.getVendor(), Triple::PC); - EXPECT_EQ(T.getOS(), Triple::Linux); + EXPECT_EQ(Triple::x86_64, T.getArch()); + EXPECT_EQ(Triple::PC, T.getVendor()); + EXPECT_EQ(Triple::Linux, T.getOS()); T = Triple("powerpc-dunno-notsure"); - EXPECT_EQ(T.getArch(), Triple::ppc); - EXPECT_EQ(T.getVendor(), Triple::UnknownVendor); - EXPECT_EQ(T.getOS(), Triple::UnknownOS); + EXPECT_EQ(Triple::ppc, T.getArch()); + EXPECT_EQ(Triple::UnknownVendor, T.getVendor()); + EXPECT_EQ(Triple::UnknownOS, T.getOS()); T = Triple("huh"); - EXPECT_EQ(T.getArch(), Triple::UnknownArch); + EXPECT_EQ(Triple::UnknownArch, T.getArch()); } TEST(TripleTest, MutateName) { Triple T; - EXPECT_EQ(T.getArch(), Triple::UnknownArch); - EXPECT_EQ(T.getVendor(), Triple::UnknownVendor); - EXPECT_EQ(T.getOS(), Triple::UnknownOS); + EXPECT_EQ(Triple::UnknownArch, T.getArch()); + EXPECT_EQ(Triple::UnknownVendor, T.getVendor()); + EXPECT_EQ(Triple::UnknownOS, T.getOS()); T.setArchName("i386"); - EXPECT_EQ(T.getArch(), Triple::x86); - EXPECT_EQ(T.getTriple(), "i386--"); + EXPECT_EQ(Triple::x86, T.getArch()); + EXPECT_EQ("i386--", T.getTriple()); T.setVendorName("pc"); - EXPECT_EQ(T.getArch(), Triple::x86); - EXPECT_EQ(T.getVendor(), Triple::PC); - EXPECT_EQ(T.getTriple(), "i386-pc-"); + EXPECT_EQ(Triple::x86, T.getArch()); + EXPECT_EQ(Triple::PC, T.getVendor()); + EXPECT_EQ("i386-pc-", T.getTriple()); T.setOSName("linux"); - EXPECT_EQ(T.getArch(), Triple::x86); - EXPECT_EQ(T.getVendor(), Triple::PC); - EXPECT_EQ(T.getOS(), Triple::Linux); - EXPECT_EQ(T.getTriple(), "i386-pc-linux"); + EXPECT_EQ(Triple::x86, T.getArch()); + EXPECT_EQ(Triple::PC, T.getVendor()); + EXPECT_EQ(Triple::Linux, T.getOS()); + EXPECT_EQ("i386-pc-linux", T.getTriple()); T.setEnvironmentName("gnu"); - EXPECT_EQ(T.getArch(), Triple::x86); - EXPECT_EQ(T.getVendor(), Triple::PC); - EXPECT_EQ(T.getOS(), Triple::Linux); - EXPECT_EQ(T.getTriple(), "i386-pc-linux-gnu"); + EXPECT_EQ(Triple::x86, T.getArch()); + EXPECT_EQ(Triple::PC, T.getVendor()); + EXPECT_EQ(Triple::Linux, T.getOS()); + EXPECT_EQ("i386-pc-linux-gnu", T.getTriple()); T.setOSName("freebsd"); - EXPECT_EQ(T.getArch(), Triple::x86); - EXPECT_EQ(T.getVendor(), Triple::PC); - EXPECT_EQ(T.getOS(), Triple::FreeBSD); - EXPECT_EQ(T.getTriple(), "i386-pc-freebsd-gnu"); + EXPECT_EQ(Triple::x86, T.getArch()); + EXPECT_EQ(Triple::PC, T.getVendor()); + EXPECT_EQ(Triple::FreeBSD, T.getOS()); + EXPECT_EQ("i386-pc-freebsd-gnu", T.getTriple()); T.setOSAndEnvironmentName("darwin"); - EXPECT_EQ(T.getArch(), Triple::x86); - EXPECT_EQ(T.getVendor(), Triple::PC); - EXPECT_EQ(T.getOS(), Triple::Darwin); - EXPECT_EQ(T.getTriple(), "i386-pc-darwin"); + EXPECT_EQ(Triple::x86, T.getArch()); + EXPECT_EQ(Triple::PC, T.getVendor()); + EXPECT_EQ(Triple::Darwin, T.getOS()); + EXPECT_EQ("i386-pc-darwin", T.getTriple()); } } diff --git a/utils/GenLibDeps.pl b/utils/GenLibDeps.pl index 73f3e71491a..6d0b13ee436 100755 --- a/utils/GenLibDeps.pl +++ b/utils/GenLibDeps.pl @@ -38,6 +38,10 @@ if (!$FLAT) { die "Can't find 'dot'" if (! -x "$DotPath"); } +if (defined($ENV{NM})) { + chomp($nmPath=$ENV{NM}); +} + if (!defined($nmPath) || $nmPath eq "") { chomp($nmPath=`which nm`); die "Can't find 'nm'" if (! -x "$nmPath"); @@ -96,7 +100,7 @@ sub gen_one_entry { print "
    $lib
      \n"; } open UNDEFS, - "$nmPath -g -u $Directory/$lib | sed -e 's/^[ 0]* U //' | sort | uniq |"; + "$nmPath -u $Directory/$lib | sed -e 's/^[ 0]* U //' | sort | uniq |"; my %DepLibs; while () { chomp; @@ -116,7 +120,7 @@ sub gen_one_entry { close UNDEFS or die "nm failed"; unless(keys %DepLibs) { # above failed - open UNDEFS, "$nmPath -g -u $Directory/$lib |"; + open UNDEFS, "$nmPath -u $Directory/$lib |"; while () { # to bypass non-working sed if (' ' eq substr($_,0,2) and index($_,'U ')) { diff --git a/utils/NewNightlyTest.pl b/utils/NewNightlyTest.pl index a40b3f11a2b..feac974be87 100755 --- a/utils/NewNightlyTest.pl +++ b/utils/NewNightlyTest.pl @@ -11,8 +11,6 @@ use Socket; # regressions and performance changes. Submits this information # to llvm.org where it is placed into the nightlytestresults database. # -# Modified heavily by Patrick Jenkins, July 2006 -# # Syntax: NightlyTest.pl [OPTIONS] [CVSROOT BUILDDIR WEBDIR] # where # OPTIONS may include one or more of the following: @@ -26,10 +24,12 @@ use Socket; # -nodejagnu Do not run feature or regression tests # -parallel Run parallel jobs with GNU Make (see -parallel-jobs). # -parallel-jobs The number of parallel Make jobs to use (default is two). +# -with-clang Checkout Clang source into tools/clang. # -release Build an LLVM Release version # -release-asserts Build an LLVM ReleaseAsserts version # -enable-llcbeta Enable testing of beta features in llc. # -enable-lli Enable testing of lli (interpreter) features, default is off +# -disable-pic Disable building with Position Independent Code. # -disable-llc Disable LLC tests in the nightly tester. # -disable-jit Disable JIT tests in the nightly tester. # -disable-cbe Disable C backend tests in the nightly tester. @@ -98,7 +98,7 @@ use Socket; ############################################################## my $HOME = $ENV{'HOME'}; my $SVNURL = $ENV{"SVNURL"}; -$SVNURL = 'https://llvm.org/svn/llvm-project' unless $SVNURL; +$SVNURL = 'http://llvm.org/svn/llvm-project' unless $SVNURL; my $CVSRootDir = $ENV{'CVSROOT'}; $CVSRootDir = "/home/vadve/shared/PublicCVS" unless $CVSRootDir; my $BuildDir = $ENV{'BUILDDIR'}; @@ -145,6 +145,7 @@ while (scalar(@ARGV) and ($_ = $ARGV[0], /^[-+]/)) { if (/^-norunningtests$/) { next; } # Backward compatibility, ignored. if (/^-parallel-jobs$/) { $PARALLELJOBS = "$ARGV[0]"; shift; next;} if (/^-parallel$/) { $MAKEOPTS = "$MAKEOPTS -j$PARALLELJOBS -l3.0"; next; } + if (/^-with-clang$/) { $WITHCLANG = 1; next; } if (/^-release$/) { $MAKEOPTS = "$MAKEOPTS ENABLE_OPTIMIZED=1 ". "OPTIMIZE_OPTION=-O2"; $BUILDTYPE="release"; next;} if (/^-release-asserts$/){ $MAKEOPTS = "$MAKEOPTS ENABLE_OPTIMIZED=1 ". @@ -152,6 +153,7 @@ while (scalar(@ARGV) and ($_ = $ARGV[0], /^[-+]/)) { "OPTIMIZE_OPTION=-O2"; $BUILDTYPE="release-asserts"; next;} if (/^-enable-llcbeta$/) { $PROGTESTOPTS .= " ENABLE_LLCBETA=1"; next; } + if (/^-disable-pic$/) { $CONFIGUREARGS .= " --enable-pic=no"; next; } if (/^-enable-lli$/) { $PROGTESTOPTS .= " ENABLE_LLI=1"; $CONFIGUREARGS .= " --enable-lli"; next; } if (/^-disable-llc$/) { $PROGTESTOPTS .= " DISABLE_LLC=1"; @@ -534,13 +536,20 @@ ChangeDir( $BuildDir, "checkout directory" ); if (!$NOCHECKOUT) { if ( $VERBOSE ) { print "CHECKOUT STAGE:\n"; } if ($USESVN) { - my $SVNCMD = "$NICE svn co $SVNURL"; - if ($VERBOSE) { - print "( time -p $SVNCMD/llvm/trunk llvm; cd llvm/projects ; " . + my $SVNCMD = "$NICE svn co --non-interactive $SVNURL"; + if ($VERBOSE) { + print "( time -p $SVNCMD/llvm/trunk llvm; cd llvm/projects ; " . + "$SVNCMD/test-suite/trunk llvm-test ) > $COLog 2>&1\n"; + } + system "( time -p $SVNCMD/llvm/trunk llvm; cd llvm/projects ; " . "$SVNCMD/test-suite/trunk llvm-test ) > $COLog 2>&1\n"; - } - system "( time -p $SVNCMD/llvm/trunk llvm; cd llvm/projects ; " . - "$SVNCMD/test-suite/trunk llvm-test ) > $COLog 2>&1\n"; + if ($WITHCLANG) { + my $SVNCMD = "$NICE svn co --non-interactive $SVNURL/cfe/trunk"; + if ($VERBOSE) { + print "( time -p cd llvm/tools ; $SVNCMD clang ) > $COLog 2>&1\n"; + } + system "( time -p cd llvm/tools ; $SVNCMD clang ) > $COLog 2>&1\n"; + } } else { my $CVSOPT = ""; $CVSOPT = "-z3" # Use compression if going over ssh. @@ -611,7 +620,7 @@ if (!$NOCVSSTATS) { if ($VERBOSE) { print "CHANGE HISTORY ANALYSIS STAGE\n"; } if ($USESVN) { - @SVNHistory = split / $BuildLog 2>&1"; if ( $VERBOSE ) { print "BUILD STAGE:\n"; + print "(time -p $NICE $MAKECMD clean) >> $BuildLog 2>&1\n"; print "(time -p $NICE $MAKECMD $MAKEOPTS) >> $BuildLog 2>&1\n"; } # Build the entire tree, capturing the output into $BuildLog + system "(time -p $NICE $MAKECMD clean) >> $BuildLog 2>&1"; system "(time -p $NICE $MAKECMD $MAKEOPTS) >> $BuildLog 2>&1"; } diff --git a/utils/TableGen/ClangDiagnosticsEmitter.cpp b/utils/TableGen/ClangDiagnosticsEmitter.cpp index 919ae9befed..a4a56983ca9 100644 --- a/utils/TableGen/ClangDiagnosticsEmitter.cpp +++ b/utils/TableGen/ClangDiagnosticsEmitter.cpp @@ -65,6 +65,12 @@ void ClangDiagsDefsEmitter::run(std::ostream &OS) { } else { OS << ", 0"; } + + // SFINAE bit + if (R.getValueAsBit("SFINAE")) + OS << ", true"; + else + OS << ", false"; OS << ")\n"; } } diff --git a/utils/TableGen/CodeGenDAGPatterns.cpp b/utils/TableGen/CodeGenDAGPatterns.cpp index db76dabb537..e668468772c 100644 --- a/utils/TableGen/CodeGenDAGPatterns.cpp +++ b/utils/TableGen/CodeGenDAGPatterns.cpp @@ -2007,9 +2007,28 @@ void CodeGenDAGPatterns::ParsePatterns() { Pattern = new TreePattern(Patterns[i], Tree, true, *this); else { std::vector Values; - for (unsigned j = 0, ee = Tree->getNumArgs(); j != ee; ++j) + RecTy *ListTy = 0; + for (unsigned j = 0, ee = Tree->getNumArgs(); j != ee; ++j) { Values.push_back(Tree->getArg(j)); - ListInit *LI = new ListInit(Values); + TypedInit *TArg = dynamic_cast(Tree->getArg(j)); + if (TArg == 0) { + cerr << "In dag: " << Tree->getAsString(); + cerr << " -- Untyped argument in pattern\n"; + assert(0 && "Untyped argument in pattern"); + } + if (ListTy != 0) { + ListTy = resolveTypes(ListTy, TArg->getType()); + if (ListTy == 0) { + cerr << "In dag: " << Tree->getAsString(); + cerr << " -- Incompatible types in pattern arguments\n"; + assert(0 && "Incompatible types in pattern arguments"); + } + } + else { + ListTy = TArg->getType(); + } + } + ListInit *LI = new ListInit(Values, new ListRecTy(ListTy)); Pattern = new TreePattern(Patterns[i], LI, true, *this); } diff --git a/utils/TableGen/Record.cpp b/utils/TableGen/Record.cpp index 45804b938cb..c62e21b3aa1 100644 --- a/utils/TableGen/Record.cpp +++ b/utils/TableGen/Record.cpp @@ -189,7 +189,12 @@ Init *ListRecTy::convertValue(ListInit *LI) { else return 0; - return new ListInit(Elements); + ListRecTy *LType = dynamic_cast(LI->getType()); + if (LType == 0) { + return 0; + } + + return new ListInit(Elements, new ListRecTy(Ty)); } Init *ListRecTy::convertValue(TypedInit *TI) { @@ -270,6 +275,57 @@ bool RecordRecTy::baseClassOf(const RecordRecTy *RHS) const { } +/// resolveTypes - Find a common type that T1 and T2 convert to. +/// Return 0 if no such type exists. +/// +RecTy *llvm::resolveTypes(RecTy *T1, RecTy *T2) { + if (!T1->typeIsConvertibleTo(T2)) { + if (!T2->typeIsConvertibleTo(T1)) { + // If one is a Record type, check superclasses + RecordRecTy *RecTy1 = dynamic_cast(T1); + if (RecTy1) { + // See if T2 inherits from a type T1 also inherits from + const std::vector &T1SuperClasses = RecTy1->getRecord()->getSuperClasses(); + for(std::vector::const_iterator i = T1SuperClasses.begin(), + iend = T1SuperClasses.end(); + i != iend; + ++i) { + RecordRecTy *SuperRecTy1 = new RecordRecTy(*i); + RecTy *NewType1 = resolveTypes(SuperRecTy1, T2); + if (NewType1 != 0) { + if (NewType1 != SuperRecTy1) { + delete SuperRecTy1; + } + return NewType1; + } + } + } + RecordRecTy *RecTy2 = dynamic_cast(T2); + if (RecTy2) { + // See if T1 inherits from a type T2 also inherits from + const std::vector &T2SuperClasses = RecTy2->getRecord()->getSuperClasses(); + for(std::vector::const_iterator i = T2SuperClasses.begin(), + iend = T2SuperClasses.end(); + i != iend; + ++i) { + RecordRecTy *SuperRecTy2 = new RecordRecTy(*i); + RecTy *NewType2 = resolveTypes(T1, SuperRecTy2); + if (NewType2 != 0) { + if (NewType2 != SuperRecTy2) { + delete SuperRecTy2; + } + return NewType2; + } + } + } + return 0; + } + return T2; + } + return T1; +} + + //===----------------------------------------------------------------------===// // Initializer implementations //===----------------------------------------------------------------------===// @@ -398,7 +454,7 @@ Init *ListInit::convertInitListSlice(const std::vector &Elements) { return 0; Vals.push_back(getElement(Elements[i])); } - return new ListInit(Vals); + return new ListInit(Vals, getType()); } Record *ListInit::getElementAsRecord(unsigned i) const { @@ -426,10 +482,20 @@ Init *ListInit::resolveReferences(Record &R, const RecordVal *RV) { } if (Changed) - return new ListInit(Resolved); + return new ListInit(Resolved, getType()); return this; } +Init *ListInit::resolveListElementReference(Record &R, const RecordVal *IRV, + unsigned Elt) { + if (Elt >= getSize()) + return 0; // Out of range reference. + Init *E = getElement(Elt); + if (!dynamic_cast(E)) // If the element is set + return E; // Replace the VarListElementInit with it. + return 0; +} + std::string ListInit::getAsString() const { std::string Result = "["; for (unsigned i = 0, e = Values.size(); i != e; ++i) { @@ -538,7 +604,7 @@ Init *UnOpInit::Fold(Record *CurRec, MultiClass *CurMultiClass) { assert(0 && "Empty list in cdr"); return 0; } - ListInit *Result = new ListInit(LHSl->begin()+1, LHSl->end()); + ListInit *Result = new ListInit(LHSl->begin()+1, LHSl->end(), LHSl->getType()); return Result; } break; @@ -553,6 +619,16 @@ Init *UnOpInit::Fold(Record *CurRec, MultiClass *CurMultiClass) { return new IntInit(0); } } + StringInit *LHSs = dynamic_cast(LHS); + if (LHSs) { + if (LHSs->getValue().empty()) { + return new IntInit(1); + } + else { + return new IntInit(0); + } + } + break; } } @@ -665,8 +741,8 @@ Init *BinOpInit::Fold(Record *CurRec, MultiClass *CurMultiClass) { if (Record *D = Records.getDef(Name)) return new DefInit(D); - cerr << "Variable not defined: '" + Name + "'\n"; - assert(0 && "Variable not found"); + cerr << "Variable not defined in !nameconcat: '" + Name + "'\n"; + assert(0 && "Variable not found in !nameconcat"); return 0; } break; @@ -848,7 +924,7 @@ static Init *ForeachHelper(Init *LHS, Init *MHS, Init *RHS, RecTy *Type, delete NewOp; } } - return new ListInit(NewList); + return new ListInit(NewList, MHSl->getType()); } } return 0; @@ -932,9 +1008,25 @@ Init *TernOpInit::Fold(Record *CurRec, MultiClass *CurMultiClass) { Init *TernOpInit::resolveReferences(Record &R, const RecordVal *RV) { Init *lhs = LHS->resolveReferences(R, RV); + + if (Opc == IF && lhs != LHS) { + IntInit *Value = dynamic_cast(lhs); + if (Value != 0) { + // Short-circuit + if (Value->getValue()) { + Init *mhs = MHS->resolveReferences(R, RV); + return (new TernOpInit(getOpcode(), lhs, mhs, RHS, getType()))->Fold(&R, 0); + } + else { + Init *rhs = RHS->resolveReferences(R, RV); + return (new TernOpInit(getOpcode(), lhs, MHS, rhs, getType()))->Fold(&R, 0); + } + } + } + Init *mhs = MHS->resolveReferences(R, RV); Init *rhs = RHS->resolveReferences(R, RV); - + if (LHS != lhs || MHS != mhs || RHS != rhs) return (new TernOpInit(getOpcode(), lhs, mhs, rhs, getType()))->Fold(&R, 0); return Fold(&R, 0); @@ -978,7 +1070,7 @@ Init *TypedInit::convertInitListSlice(const std::vector &Elements) { ListInits.reserve(Elements.size()); for (unsigned i = 0, e = Elements.size(); i != e; ++i) ListInits.push_back(new VarListElementInit(this, Elements[i])); - return new ListInit(ListInits); + return new ListInit(ListInits, T); } diff --git a/utils/TableGen/Record.h b/utils/TableGen/Record.h index 4284cabf855..ac06cae809d 100644 --- a/utils/TableGen/Record.h +++ b/utils/TableGen/Record.h @@ -442,7 +442,10 @@ public: virtual bool baseClassOf(const RecordRecTy *RHS) const; }; - +/// resolveTypes - Find a common type that T1 and T2 convert to. +/// Return 0 if no such type exists. +/// +RecTy *resolveTypes(RecTy *T1, RecTy *T2); //===----------------------------------------------------------------------===// // Initializer Classes @@ -618,10 +621,10 @@ public: /// IntInit - 7 - Represent an initalization by a literal integer value. /// -class IntInit : public Init { +class IntInit : public TypedInit { int64_t Value; public: - explicit IntInit(int64_t V) : Value(V) {} + explicit IntInit(int64_t V) : TypedInit(new IntRecTy), Value(V) {} int64_t getValue() const { return Value; } @@ -631,6 +634,25 @@ public: virtual Init *convertInitializerBitRange(const std::vector &Bits); virtual std::string getAsString() const; + + /// resolveBitReference - This method is used to implement + /// VarBitInit::resolveReferences. If the bit is able to be resolved, we + /// simply return the resolved value, otherwise we return null. + /// + virtual Init *resolveBitReference(Record &R, const RecordVal *RV, + unsigned Bit) { + assert(0 && "Illegal bit reference off int"); + return 0; + } + + /// resolveListElementReference - This method is used to implement + /// VarListElementInit::resolveReferences. If the list element is resolvable + /// now, we return the resolved value, otherwise we return null. + virtual Init *resolveListElementReference(Record &R, const RecordVal *RV, + unsigned Elt) { + assert(0 && "Illegal element reference off int"); + return 0; + } }; @@ -688,17 +710,18 @@ public: /// ListInit - [AL, AH, CL] - Represent a list of defs /// -class ListInit : public Init { +class ListInit : public TypedInit { std::vector Values; public: typedef std::vector::iterator iterator; typedef std::vector::const_iterator const_iterator; - explicit ListInit(std::vector &Vs) { + explicit ListInit(std::vector &Vs, RecTy *EltTy) + : TypedInit(new ListRecTy(EltTy)) { Values.swap(Vs); } - explicit ListInit(iterator Start, iterator End) - : Values(Start, End) {} + explicit ListInit(iterator Start, iterator End, RecTy *EltTy) + : TypedInit(new ListRecTy(EltTy)), Values(Start, End) {} unsigned getSize() const { return Values.size(); } Init *getElement(unsigned i) const { @@ -730,6 +753,22 @@ public: inline size_t size () const { return Values.size(); } inline bool empty() const { return Values.empty(); } + + /// resolveBitReference - This method is used to implement + /// VarBitInit::resolveReferences. If the bit is able to be resolved, we + /// simply return the resolved value, otherwise we return null. + /// + virtual Init *resolveBitReference(Record &R, const RecordVal *RV, + unsigned Bit) { + assert(0 && "Illegal bit reference off list"); + return 0; + } + + /// resolveListElementReference - This method is used to implement + /// VarListElementInit::resolveReferences. If the list element is resolvable + /// now, we return the resolved value, otherwise we return null. + virtual Init *resolveListElementReference(Record &R, const RecordVal *RV, + unsigned Elt); }; diff --git a/utils/TableGen/TGParser.cpp b/utils/TableGen/TGParser.cpp index fc6f29fd9f1..cdd285703be 100644 --- a/utils/TableGen/TGParser.cpp +++ b/utils/TableGen/TGParser.cpp @@ -12,6 +12,7 @@ //===----------------------------------------------------------------------===// #include +#include #include "TGParser.h" #include "Record.h" @@ -396,7 +397,7 @@ ParseSubClassReference(Record *CurRec, bool isDefm) { return Result; } - Result.TemplateArgs = ParseValueList(CurRec); + Result.TemplateArgs = ParseValueList(CurRec, Result.Rec); if (Result.TemplateArgs.empty()) { Result.Rec = 0; // Error parsing value list. return Result; @@ -438,7 +439,7 @@ ParseSubMultiClassReference(MultiClass *CurMC) { return Result; } - Result.TemplateArgs = ParseValueList(&CurMC->Rec); + Result.TemplateArgs = ParseValueList(&CurMC->Rec, &Result.MC->Rec); if (Result.TemplateArgs.empty()) { Result.MC = 0; // Error parsing value list. return Result; @@ -728,21 +729,28 @@ Init *TGParser::ParseOperation(Record *CurRec) { || Code == UnOpInit::CDR || Code == UnOpInit::LNULL) { ListInit *LHSl = dynamic_cast(LHS); + StringInit *LHSs = dynamic_cast(LHS); TypedInit *LHSt = dynamic_cast(LHS); - if (LHSl == 0 && LHSt == 0) { - TokError("expected list type argument in unary operator"); + if (LHSl == 0 && LHSs == 0 && LHSt == 0) { + TokError("expected list or string type argument in unary operator"); return 0; } if (LHSt) { ListRecTy *LType = dynamic_cast(LHSt->getType()); - if (LType == 0) { - TokError("expected list type argumnet in unary operator"); + StringRecTy *SType = dynamic_cast(LHSt->getType()); + if (LType == 0 && SType == 0) { + TokError("expected list or string type argumnet in unary operator"); return 0; } } if (Code == UnOpInit::CAR || Code == UnOpInit::CDR) { + if (LHSl == 0 && LHSt == 0) { + TokError("expected list type argumnet in unary operator"); + return 0; + } + if (LHSl && LHSl->getSize() == 0) { TokError("empty list argument in unary operator"); return 0; @@ -1011,7 +1019,7 @@ RecTy *TGParser::ParseOperatorType(void) { /// SimpleValue ::= SRLTOK '(' Value ',' Value ')' /// SimpleValue ::= STRCONCATTOK '(' Value ',' Value ')' /// -Init *TGParser::ParseSimpleValue(Record *CurRec) { +Init *TGParser::ParseSimpleValue(Record *CurRec, RecTy *ItemType) { Init *R = 0; switch (Lex.getCode()) { default: TokError("Unknown token when parsing a value"); break; @@ -1043,15 +1051,7 @@ Init *TGParser::ParseSimpleValue(Record *CurRec) { TokError("expected non-empty value list"); return 0; } - std::vector ValueList = ParseValueList(CurRec); - if (ValueList.empty()) return 0; - - if (Lex.getCode() != tgtok::greater) { - TokError("expected '>' at end of value list"); - return 0; - } - Lex.Lex(); // eat the '>' - + // This is a CLASS expression. This is supposed to synthesize // a new anonymous definition, deriving from CLASS with no // body. @@ -1060,6 +1060,15 @@ Init *TGParser::ParseSimpleValue(Record *CurRec) { Error(NameLoc, "Expected a class name, got '" + Name + "'"); return 0; } + + std::vector ValueList = ParseValueList(CurRec, Class); + if (ValueList.empty()) return 0; + + if (Lex.getCode() != tgtok::greater) { + TokError("expected '>' at end of value list"); + return 0; + } + Lex.Lex(); // eat the '>' // Create the new record, set it as CurRec temporarily. static unsigned AnonCounter = 0; @@ -1108,8 +1117,22 @@ Init *TGParser::ParseSimpleValue(Record *CurRec) { Lex.Lex(); // eat the '[' std::vector Vals; + RecTy *DeducedEltTy = 0; + ListRecTy *GivenListTy = 0; + + if (ItemType != 0) { + ListRecTy *ListType = dynamic_cast(ItemType); + if (ListType == 0) { + std::stringstream s; + s << "Type mismatch for list, expected list type, got " + << ItemType->getAsString(); + TokError(s.str()); + } + GivenListTy = ListType; + } + if (Lex.getCode() != tgtok::r_square) { - Vals = ParseValueList(CurRec); + Vals = ParseValueList(CurRec, 0, GivenListTy ? GivenListTy->getElementType() : 0); if (Vals.empty()) return 0; } if (Lex.getCode() != tgtok::r_square) { @@ -1117,7 +1140,77 @@ Init *TGParser::ParseSimpleValue(Record *CurRec) { return 0; } Lex.Lex(); // eat the ']' - return new ListInit(Vals); + + RecTy *GivenEltTy = 0; + if (Lex.getCode() == tgtok::less) { + // Optional list element type + Lex.Lex(); // eat the '<' + + GivenEltTy = ParseType(); + if (GivenEltTy == 0) { + // Couldn't parse element type + return 0; + } + + if (Lex.getCode() != tgtok::greater) { + TokError("expected '>' at end of list element type"); + return 0; + } + Lex.Lex(); // eat the '>' + } + + // Check elements + RecTy *EltTy = 0; + for (std::vector::iterator i = Vals.begin(), ie = Vals.end(); + i != ie; + ++i) { + TypedInit *TArg = dynamic_cast(*i); + if (TArg == 0) { + TokError("Untyped list element"); + return 0; + } + if (EltTy != 0) { + EltTy = resolveTypes(EltTy, TArg->getType()); + if (EltTy == 0) { + TokError("Incompatible types in list elements"); + return 0; + } + } + else { + EltTy = TArg->getType(); + } + } + + if (GivenEltTy != 0) { + if (EltTy != 0) { + // Verify consistency + if (!EltTy->typeIsConvertibleTo(GivenEltTy)) { + TokError("Incompatible types in list elements"); + return 0; + } + } + EltTy = GivenEltTy; + } + + if (EltTy == 0) { + if (ItemType == 0) { + TokError("No type for list"); + return 0; + } + DeducedEltTy = GivenListTy->getElementType(); + } + else { + // Make sure the deduced type is compatible with the given type + if (GivenListTy) { + if (!EltTy->typeIsConvertibleTo(GivenListTy->getElementType())) { + TokError("Element type mismatch for list"); + return 0; + } + } + DeducedEltTy = EltTy; + } + + return new ListInit(Vals, DeducedEltTy); } case tgtok::l_paren: { // Value ::= '(' IDValue DagArgList ')' Lex.Lex(); // eat the '(' @@ -1193,8 +1286,8 @@ Init *TGParser::ParseSimpleValue(Record *CurRec) { /// ValueSuffix ::= '[' BitList ']' /// ValueSuffix ::= '.' ID /// -Init *TGParser::ParseValue(Record *CurRec) { - Init *Result = ParseSimpleValue(CurRec); +Init *TGParser::ParseValue(Record *CurRec, RecTy *ItemType) { + Init *Result = ParseSimpleValue(CurRec, ItemType); if (Result == 0) return 0; // Parse the suffixes now if present. @@ -1299,15 +1392,31 @@ TGParser::ParseDagArgList(Record *CurRec) { /// /// ValueList ::= Value (',' Value) /// -std::vector TGParser::ParseValueList(Record *CurRec) { +std::vector TGParser::ParseValueList(Record *CurRec, Record *ArgsRec, RecTy *EltTy) { std::vector Result; - Result.push_back(ParseValue(CurRec)); + RecTy *ItemType = EltTy; + int ArgN = 0; + if (ArgsRec != 0 && EltTy == 0) { + const std::vector &TArgs = ArgsRec->getTemplateArgs(); + const RecordVal *RV = ArgsRec->getValue(TArgs[ArgN]); + assert(RV && "Template argument record not found??"); + ItemType = RV->getType(); + ++ArgN; + } + Result.push_back(ParseValue(CurRec, ItemType)); if (Result.back() == 0) return std::vector(); while (Lex.getCode() == tgtok::comma) { Lex.Lex(); // Eat the comma - Result.push_back(ParseValue(CurRec)); + if (ArgsRec != 0 && EltTy == 0) { + const std::vector &TArgs = ArgsRec->getTemplateArgs(); + const RecordVal *RV = ArgsRec->getValue(TArgs[ArgN]); + assert(RV && "Template argument record not found??"); + ItemType = RV->getType(); + ++ArgN; + } + Result.push_back(ParseValue(CurRec, ItemType)); if (Result.back() == 0) return std::vector(); } @@ -1362,7 +1471,7 @@ std::string TGParser::ParseDeclaration(Record *CurRec, if (Lex.getCode() == tgtok::equal) { Lex.Lex(); TGLoc ValLoc = Lex.getLoc(); - Init *Val = ParseValue(CurRec); + Init *Val = ParseValue(CurRec, Type); if (Val == 0 || SetValue(CurRec, ValLoc, DeclName, std::vector(), Val)) return ""; @@ -1440,7 +1549,13 @@ bool TGParser::ParseBodyItem(Record *CurRec) { return TokError("expected '=' in let expression"); Lex.Lex(); // eat the '='. - Init *Val = ParseValue(CurRec); + RecordVal *Field = CurRec->getValue(FieldName); + if (Field == 0) + return TokError("Value '" + FieldName + "' unknown!"); + + RecTy *Type = Field->getType(); + + Init *Val = ParseValue(CurRec, Type); if (Val == 0) return true; if (Lex.getCode() != tgtok::semi) diff --git a/utils/TableGen/TGParser.h b/utils/TableGen/TGParser.h index f03052eb799..3af467d16ab 100644 --- a/utils/TableGen/TGParser.h +++ b/utils/TableGen/TGParser.h @@ -93,9 +93,9 @@ private: // Parser methods. Init *ParseIDValue(Record *CurRec); Init *ParseIDValue(Record *CurRec, const std::string &Name, TGLoc NameLoc); - Init *ParseSimpleValue(Record *CurRec); - Init *ParseValue(Record *CurRec); - std::vector ParseValueList(Record *CurRec); + Init *ParseSimpleValue(Record *CurRec, RecTy *ItemType = 0); + Init *ParseValue(Record *CurRec, RecTy *ItemType = 0); + std::vector ParseValueList(Record *CurRec, Record *ArgsRec = 0, RecTy *EltTy = 0); std::vector > ParseDagArgList(Record *); bool ParseOptionalRangeList(std::vector &Ranges); bool ParseOptionalBitList(std::vector &Ranges); diff --git a/utils/crosstool/ARM/build-install-linux.sh b/utils/crosstool/ARM/build-install-linux.sh new file mode 100755 index 00000000000..33833b5612a --- /dev/null +++ b/utils/crosstool/ARM/build-install-linux.sh @@ -0,0 +1,203 @@ +#!/bin/bash +# +# Compiles and installs a Linux/x86_64 -> Linux/ARM crosstool based on LLVM and +# LLVM-GCC-4.2 using SVN snapshots in provided tarballs. + +set -o nounset +set -o errexit + +echo -n "Welcome to LLVM Linux/X86_64 -> Linux/ARM crosstool " +echo "builder/installer; some steps will require sudo privileges." + +readonly INSTALL_ROOT="${INSTALL_ROOT:-/usr/local}" +# Both $USER and root *must* have read/write access to this dir. +readonly SCRATCH_ROOT=$(mktemp -d "${TMPDIR:-/tmp}/llvm-project.XXXXXX") +readonly SRC_ROOT="${SCRATCH_ROOT}/src" +readonly OBJ_ROOT="${SCRATCH_ROOT}/obj" + +readonly CROSS_HOST="x86_64-unknown-linux-gnu" +readonly CROSS_TARGET="arm-none-linux-gnueabi" + +readonly CODE_SOURCERY="${INSTALL_ROOT}/codesourcery" +readonly CODE_SOURCERY_PKG_PATH="${CODE_SOURCERY_PKG_PATH:-${HOME}/codesourcery}" +readonly CODE_SOURCERY_HTTP="http://www.codesourcery.com/sgpp/lite/arm/portal/package1787/public" +readonly CODE_SOURCERY_PKG="arm-2007q3-51-arm-none-linux-gnueabi-i686-pc-linux-gnu.tar.bz2" +readonly CODE_SOURCERY_ROOT="${CODE_SOURCERY}/arm-2007q3" +readonly CODE_SOURCERY_BIN="${CODE_SOURCERY_ROOT}/bin" +# Make sure ${CROSS_TARGET}-* binutils are in command path +export PATH="${CODE_SOURCERY_BIN}:${PATH}" + +readonly CROSS_TARGET_AS="${CODE_SOURCERY_BIN}/${CROSS_TARGET}-as" +readonly CROSS_TARGET_LD="${CODE_SOURCERY_BIN}/${CROSS_TARGET}-ld" + +readonly SYSROOT="${CODE_SOURCERY_ROOT}/${CROSS_TARGET}/libc" + +readonly LLVM_PROJECT="${INSTALL_ROOT}/llvm-project" +readonly LLVM_INSTALL_ROOT="${LLVM_PROJECT}/${CROSS_HOST}/${CROSS_TARGET}" +readonly LLVM_PKG_PATH="${LLVM_PKG_PATH:-${HOME}/llvm-project/snapshots}" + +# Latest SVN revision known to be working in this configuration. +readonly LLVM_DEFAULT_REV="70786" + +readonly LLVM_PKG="llvm-${LLVM_SVN_REV:-${LLVM_DEFAULT_REV}}.tar.bz2" +readonly LLVM_SRC_DIR="${SRC_ROOT}/llvm" +readonly LLVM_OBJ_DIR="${OBJ_ROOT}/llvm" +readonly LLVM_INSTALL_DIR="${LLVM_INSTALL_ROOT}/llvm" + +readonly LLVMGCC_PKG="llvm-gcc-4.2-${LLVMGCC_SVN_REV:-${LLVM_DEFAULT_REV}}.tar.bz2" +readonly LLVMGCC_SRC_DIR="${SRC_ROOT}/llvm-gcc-4.2" +readonly LLVMGCC_OBJ_DIR="${OBJ_ROOT}/llvm-gcc-4.2" +readonly LLVMGCC_INSTALL_DIR="${LLVM_INSTALL_ROOT}/llvm-gcc-4.2" + +readonly MAKE_OPTS="-j2" + +# Verify we aren't going to install into an existing directory as this might +# create problems as we won't have a clean install. +verifyNotDir() { + if [[ -d $1 ]]; then + echo "Install dir $1 already exists; remove it to continue." + exit + fi +} + +# Params: +# $1: directory to be created +# $2: optional mkdir command prefix, e.g. "sudo" +createDir() { + if [[ ! -e $1 ]]; then + ${2:-} mkdir -p $1 + elif [[ -e $1 && ! -d $1 ]]; then + echo "$1 exists but is not a directory; exiting." + exit 3 + fi +} + +sudoCreateDir() { + createDir $1 sudo + sudo chown ${USER} $1 +} + +# Prints out and runs the command, but without logging -- intended for use with +# lightweight commands that don't have useful output to parse, e.g. mkdir, tar, +# etc. +runCommand() { + local message="$1" + shift + echo "=> $message" + echo "==> Running: $*" + $* +} + +runAndLog() { + local message="$1" + local log_file="$2" + shift 2 + echo "=> $message; log in $log_file" + echo "==> Running: $*" + # Pop-up a terminal with the output of the current command? + # e.g.: xterm -e /bin/bash -c "$* >| tee $log_file" + $* &> $log_file + if [[ $? != 0 ]]; then + echo "Error occurred: see most recent log file for details." + exit + fi +} + +installCodeSourcery() { + # Create CodeSourcery dir, if necessary. + verifyNotDir ${CODE_SOURCERY} + sudoCreateDir ${CODE_SOURCERY} + + # Unpack the tarball. + if [[ ! -d ${CODE_SOURCERY_ROOT} ]]; then + cd ${CODE_SOURCERY} + if [[ -e ${CODE_SOURCERY_PKG_PATH}/${CODE_SOURCERY_PKG} ]]; then + runCommand "Unpacking CodeSourcery in ${CODE_SOURCERY}" \ + tar jxf ${CODE_SOURCERY_PKG_PATH}/${CODE_SOURCERY_PKG} + else + echo -n "CodeSourcery tarball not found in " + echo "${CODE_SOURCERY_PKG_PATH}/${CODE_SOURCERY_PKG}" + echo -n "Fix the path or download it from " + echo "${CODE_SOURCERY_HTTP}/${CROSS_TARGET}/${CODE_SOURCERY_PKG}" + exit + fi + else + echo "CodeSourcery install dir already exists." + fi + + # Verify our CodeSourcery toolchain installation. + if [[ ! -d "${SYSROOT}" ]]; then + echo -n "Error: CodeSourcery does not contain libc for ${CROSS_TARGET}: " + echo "${SYSROOT} not found." + exit + fi + + for tool in ${CROSS_TARGET_AS} ${CROSS_TARGET_LD}; do + if [[ ! -e $tool ]]; then + echo "${tool} not found; exiting." + exit + fi + done +} + +installLLVM() { + verifyNotDir ${LLVM_INSTALL_DIR} + sudoCreateDir ${LLVM_INSTALL_DIR} + + # Unpack LLVM tarball; should create the directory "llvm". + cd ${SRC_ROOT} + runCommand "Unpacking LLVM" tar jxf ${LLVM_PKG_PATH}/${LLVM_PKG} + + # Configure, build, and install LLVM. + createDir ${LLVM_OBJ_DIR} + cd ${LLVM_OBJ_DIR} + runAndLog "Configuring LLVM" ${LLVM_OBJ_DIR}/llvm-configure.log \ + ${LLVM_SRC_DIR}/configure \ + --disable-jit \ + --enable-optimized \ + --prefix=${LLVM_INSTALL_DIR} \ + --target=${CROSS_TARGET} \ + --with-llvmgccdir=${LLVMGCC_INSTALL_DIR} + runAndLog "Building LLVM" ${LLVM_OBJ_DIR}/llvm-build.log \ + make ${MAKE_OPTS} + runAndLog "Installing LLVM" ${LLVM_OBJ_DIR}/llvm-install.log \ + make ${MAKE_OPTS} install +} + +installLLVMGCC() { + verifyNotDir ${LLVMGCC_INSTALL_DIR} + sudoCreateDir ${LLVMGCC_INSTALL_DIR} + + # Unpack LLVM-GCC tarball; should create the directory "llvm-gcc-4.2". + cd ${SRC_ROOT} + runCommand "Unpacking LLVM-GCC" tar jxf ${LLVM_PKG_PATH}/${LLVMGCC_PKG} + + # Configure, build, and install LLVM-GCC. + createDir ${LLVMGCC_OBJ_DIR} + cd ${LLVMGCC_OBJ_DIR} + runAndLog "Configuring LLVM-GCC" ${LLVMGCC_OBJ_DIR}/llvmgcc-configure.log \ + ${LLVMGCC_SRC_DIR}/configure \ + --enable-languages=c,c++ \ + --enable-llvm=${LLVM_INSTALL_DIR} \ + --prefix=${LLVMGCC_INSTALL_DIR} \ + --program-prefix=llvm- \ + --target=${CROSS_TARGET} \ + --with-gnu-as=${CROSS_TARGET_AS} \ + --with-gnu-ld=${CROSS_TARGET_LD} \ + --with-sysroot=${SYSROOT} + runAndLog "Building LLVM-GCC" ${LLVMGCC_OBJ_DIR}/llvmgcc-build.log \ + make + runAndLog "Installing LLVM-GCC" ${LLVMGCC_OBJ_DIR}/llvmgcc-install.log \ + make install +} + +echo "Building in ${SCRATCH_ROOT}; installing in ${INSTALL_ROOT}" + +createDir ${SRC_ROOT} +createDir ${OBJ_ROOT} + +installCodeSourcery +installLLVM +installLLVMGCC + +echo "Done." diff --git a/utils/crosstool/create-snapshots.sh b/utils/crosstool/create-snapshots.sh new file mode 100755 index 00000000000..7c640bc00e7 --- /dev/null +++ b/utils/crosstool/create-snapshots.sh @@ -0,0 +1,41 @@ +#!/bin/bash +# +# Creates LLVM SVN snapshots: llvm-$REV.tar.bz2 and llvm-gcc-4.2-$REV.tar.bz2, +# where $REV is an SVN revision of LLVM. This is used for creating stable +# tarballs which can be used to build known-to-work crosstools. +# +# Syntax: +# $0 [REV] -- grabs the revision $REV from SVN; if not specified, grabs the +# latest SVN revision. + +set -o nounset +set -o errexit + +readonly REV="${1:-HEAD}" + +runOnModule() { + local module=$1 + local log="${module}.log" + echo "Running: svn co -r ${REV} ${module}; log in ${log}" + svn co -r ${REV} http://llvm.org/svn/llvm-project/${module}/trunk ${module} \ + > ${log} 2>&1 + + # Delete all the ".svn" dirs; they take quite a lot of space. + echo "Cleaning up .svn dirs" + find ${module} -type d -name \.svn -print0 | xargs -0 /bin/rm -rf + + # Create "module-revision.tar.bz2" packages from the SVN checkout dirs. + local revision=$(grep "Checked out revision" ${log} | \ + sed 's/[^0-9]\+\([0-9]\+\)[^0-9]\+/\1/') + local tarball="${module}-${revision}.tar.bz2" + echo "Creating tarball: ${tarball}" + tar cjf ${tarball} ${module} + + echo "Cleaning SVN checkout dir ${module}" + rm -rf ${module} ${log} +} + +for module in "llvm" "llvm-gcc-4.2"; do + runOnModule ${module} +done +