libc/src/__support/FPUtil/FPBits.h:111:14: error: no matching function for call to bit_cast

[petrhosek]

I encountered the following error when trying to compile LLVM libc for baremetal riscv32 target:

/usr/local/google/home/phosek/llvm/llvm-project/build/baremetal/./bin/clang++ --target=riscv32-unknown-elf -DLIBC_NAMESPACE=__llvm_libc_18_0_0_git -D_DEBUG -D_FILE_OFFSET_BITS=64 -D_GLIBCXX_ASSERTIONS -D_LARGEFILE_SOURCE -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -I/usr/local/google/home/phosek/llvm/llvm-project/libc -isystem /usr/local/google/home/phosek/llvm/llvm-project/build/baremetal/include
/riscv32-unknown-elf --target=riscv32-unknown-elf -march=rv32imafc -mabi=ilp32f -fPIC -fno-semantic-interposition -fvisibility-inlines-hidden -Werror=date-time -Werror=unguarded-availability-new -Wall -Wextra -Wno-unused-parameter -Wwrite-strings -Wcast-qual -Wmissing-field-initializers -Wimplicit-fallthrough -Wcovered-switch-default -Wno-noexcept-type -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wsuggest-override -Wstrin
g-conversion -Wmisleading-indentation -Wctad-maybe-unsupported -ffunction-sections -fdata-sections -ffile-prefix-map=/usr/local/google/home/phosek/llvm/llvm-project/build/baremetal/runtimes/runtimes-riscv32-unknown-elf-bins=../../../../ -ffile-prefix-map=/usr/local/google/home/phosek/llvm/llvm-project/= -no-canonical-prefixes -O2 -g -DNDEBUG -std=gnu++17 --target=riscv32-unknown-elf -O2 -fpie -ffreestanding -fno-builtin -f
no-exceptions -fno-lax-vector-conversions -fno-unwind-tables -fno-asynchronous-unwind-tables -fno-rtti -Wall -Wextra -Wconversion -Wno-sign-conversion -Wimplicit-fallthrough -Wwrite-strings -Wextra-semi -Wnewline-eof -Wnonportable-system-include-path -Wstrict-prototypes -Wthread-safety -Wglobal-constructors -DLIBC_COPT_PUBLIC_PACKAGING -UNDEBUG -MD -MT libc/src/math/generic/CMakeFiles/libc.src.math.generic.fabsl.dir/fabsl.
cpp.obj -MF libc/src/math/generic/CMakeFiles/libc.src.math.generic.fabsl.dir/fabsl.cpp.obj.d -o libc/src/math/generic/CMakeFiles/libc.src.math.generic.fabsl.dir/fabsl.cpp.obj -c /usr/local/google/home/phosek/llvm/llvm-project/libc/src/math/generic/fabsl.cpp                                                                                                                                                                         
In file included from /usr/local/google/home/phosek/llvm/llvm-project/libc/src/math/generic/fabsl.cpp:10:                                                                                                            
In file included from /usr/local/google/home/phosek/llvm/llvm-project/libc/src/__support/FPUtil/BasicOperations.h:12:                                                                                                                                                                                                                                                                                                                     
/usr/local/google/home/phosek/llvm/llvm-project/libc/src/__support/FPUtil/FPBits.h:111:14: error: no matching function for call to 'bit_cast'                                                                        
  111 |       : bits(cpp::bit_cast<UIntType>(x)) {}                                                       
      |              ^~~~~~~~~~~~~~~~~~~~~~~                                                                                                                                                                                                                                                                                                                                                                                              
/usr/local/google/home/phosek/llvm/llvm-project/libc/src/__support/FPUtil/BasicOperations.h:22:13: note: in instantiation of function template specialization '__llvm_libc_18_0_0_git::fputil::FPBits<long double>::FPBits<long double, 0>' requested here                                                                                                                                                                                
   22 |   FPBits<T> bits(x);                                                                              
      |             ^                                                                                     
/usr/local/google/home/phosek/llvm/llvm-project/libc/src/math/generic/fabsl.cpp:16:18: note: in instantiation of function template specialization '__llvm_libc_18_0_0_git::fputil::abs<long double, 0>' requested here                                                                                                                                                                                                                    
   16 |   return fputil::abs(x);                                                                                                                                                                                     
      |                  ^                                                                                                                                                                                           
/usr/local/google/home/phosek/llvm/llvm-project/libc/src/__support/CPP/bit.h:36:26: note: candidate template ignored: substitution failure [with To = UIntType, From = long double, $2 = cpp::enable_if_t<sizeof(UIntType) == sizeof(long double)>]: implicit instantiation of undefined template '__llvm_libc_18_0_0_git::cpp::enable_if<false>'                                                                                         
   36 | LIBC_INLINE constexpr To bit_cast(const From &from) {                                                                                                                                                        
      |                          ^                                                                                                                                                                                                                                                                                                                                                                                                        
1 error generated.

[llvmbot]

@llvm/issue-subscribers-libc

Author: Petr Hosek (petrhosek)

I encountered the following error when trying to compile LLVM libc for baremetal riscv32 target: ``` /usr/local/google/home/phosek/llvm/llvm-project/build/baremetal/./bin/clang++ --target=riscv32-unknown-elf -DLIBC_NAMESPACE=__llvm_libc_18_0_0_git -D_DEBUG -D_FILE_OFFSET_BITS=64 -D_GLIBCXX_ASSERTIONS -D_LARGEFILE_SOURCE -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -I/usr/local/google/home/phosek/llvm/llvm-project/libc -isystem /usr/local/google/home/phosek/llvm/llvm-project/build/baremetal/include /riscv32-unknown-elf --target=riscv32-unknown-elf -march=rv32imafc -mabi=ilp32f -fPIC -fno-semantic-interposition -fvisibility-inlines-hidden -Werror=date-time -Werror=unguarded-availability-new -Wall -Wextra -Wno-unused-parameter -Wwrite-strings -Wcast-qual -Wmissing-field-initializers -Wimplicit-fallthrough -Wcovered-switch-default -Wno-noexcept-type -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wsuggest-override -Wstrin g-conversion -Wmisleading-indentation -Wctad-maybe-unsupported -ffunction-sections -fdata-sections -ffile-prefix-map=/usr/local/google/home/phosek/llvm/llvm-project/build/baremetal/runtimes/runtimes-riscv32-unknown-elf-bins=../../../../ -ffile-prefix-map=/usr/local/google/home/phosek/llvm/llvm-project/= -no-canonical-prefixes -O2 -g -DNDEBUG -std=gnu++17 --target=riscv32-unknown-elf -O2 -fpie -ffreestanding -fno-builtin -f no-exceptions -fno-lax-vector-conversions -fno-unwind-tables -fno-asynchronous-unwind-tables -fno-rtti -Wall -Wextra -Wconversion -Wno-sign-conversion -Wimplicit-fallthrough -Wwrite-strings -Wextra-semi -Wnewline-eof -Wnonportable-system-include-path -Wstrict-prototypes -Wthread-safety -Wglobal-constructors -DLIBC_COPT_PUBLIC_PACKAGING -UNDEBUG -MD -MT libc/src/math/generic/CMakeFiles/libc.src.math.generic.fabsl.dir/fabsl. cpp.obj -MF libc/src/math/generic/CMakeFiles/libc.src.math.generic.fabsl.dir/fabsl.cpp.obj.d -o libc/src/math/generic/CMakeFiles/libc.src.math.generic.fabsl.dir/fabsl.cpp.obj -c /usr/local/google/home/phosek/llvm/llvm-project/libc/src/math/generic/fabsl.cpp In file included from /usr/local/google/home/phosek/llvm/llvm-project/libc/src/math/generic/fabsl.cpp:10: In file included from /usr/local/google/home/phosek/llvm/llvm-project/libc/src/__support/FPUtil/BasicOperations.h:12: /usr/local/google/home/phosek/llvm/llvm-project/libc/src/__support/FPUtil/FPBits.h:111:14: error: no matching function for call to 'bit_cast' 111 | : bits(cpp::bit_cast<UIntType>(x)) {} | ^~~~~~~~~~~~~~~~~~~~~~~ /usr/local/google/home/phosek/llvm/llvm-project/libc/src/__support/FPUtil/BasicOperations.h:22:13: note: in instantiation of function template specialization '__llvm_libc_18_0_0_git::fputil::FPBits<long double>::FPBits<long double, 0>' requested here 22 | FPBits<T> bits(x); | ^ /usr/local/google/home/phosek/llvm/llvm-project/libc/src/math/generic/fabsl.cpp:16:18: note: in instantiation of function template specialization '__llvm_libc_18_0_0_git::fputil::abs<long double, 0>' requested here 16 | return fputil::abs(x); | ^ /usr/local/google/home/phosek/llvm/llvm-project/libc/src/__support/CPP/bit.h:36:26: note: candidate template ignored: substitution failure [with To = UIntType, From = long double, $2 = cpp::enable_if_t<sizeof(UIntType) == sizeof(long double)>]: implicit instantiation of undefined template '__llvm_libc_18_0_0_git::cpp::enable_if<false>' 36 | LIBC_INLINE constexpr To bit_cast(const From &from) { | ^ 1 error generated. ```

[llvmbot]

@llvm/issue-subscribers-backend-risc-v

Author: Petr Hosek (petrhosek)

I encountered the following error when trying to compile LLVM libc for baremetal riscv32 target: ``` /usr/local/google/home/phosek/llvm/llvm-project/build/baremetal/./bin/clang++ --target=riscv32-unknown-elf -DLIBC_NAMESPACE=__llvm_libc_18_0_0_git -D_DEBUG -D_FILE_OFFSET_BITS=64 -D_GLIBCXX_ASSERTIONS -D_LARGEFILE_SOURCE -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -I/usr/local/google/home/phosek/llvm/llvm-project/libc -isystem /usr/local/google/home/phosek/llvm/llvm-project/build/baremetal/include /riscv32-unknown-elf --target=riscv32-unknown-elf -march=rv32imafc -mabi=ilp32f -fPIC -fno-semantic-interposition -fvisibility-inlines-hidden -Werror=date-time -Werror=unguarded-availability-new -Wall -Wextra -Wno-unused-parameter -Wwrite-strings -Wcast-qual -Wmissing-field-initializers -Wimplicit-fallthrough -Wcovered-switch-default -Wno-noexcept-type -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wsuggest-override -Wstrin g-conversion -Wmisleading-indentation -Wctad-maybe-unsupported -ffunction-sections -fdata-sections -ffile-prefix-map=/usr/local/google/home/phosek/llvm/llvm-project/build/baremetal/runtimes/runtimes-riscv32-unknown-elf-bins=../../../../ -ffile-prefix-map=/usr/local/google/home/phosek/llvm/llvm-project/= -no-canonical-prefixes -O2 -g -DNDEBUG -std=gnu++17 --target=riscv32-unknown-elf -O2 -fpie -ffreestanding -fno-builtin -f no-exceptions -fno-lax-vector-conversions -fno-unwind-tables -fno-asynchronous-unwind-tables -fno-rtti -Wall -Wextra -Wconversion -Wno-sign-conversion -Wimplicit-fallthrough -Wwrite-strings -Wextra-semi -Wnewline-eof -Wnonportable-system-include-path -Wstrict-prototypes -Wthread-safety -Wglobal-constructors -DLIBC_COPT_PUBLIC_PACKAGING -UNDEBUG -MD -MT libc/src/math/generic/CMakeFiles/libc.src.math.generic.fabsl.dir/fabsl. cpp.obj -MF libc/src/math/generic/CMakeFiles/libc.src.math.generic.fabsl.dir/fabsl.cpp.obj.d -o libc/src/math/generic/CMakeFiles/libc.src.math.generic.fabsl.dir/fabsl.cpp.obj -c /usr/local/google/home/phosek/llvm/llvm-project/libc/src/math/generic/fabsl.cpp In file included from /usr/local/google/home/phosek/llvm/llvm-project/libc/src/math/generic/fabsl.cpp:10: In file included from /usr/local/google/home/phosek/llvm/llvm-project/libc/src/__support/FPUtil/BasicOperations.h:12: /usr/local/google/home/phosek/llvm/llvm-project/libc/src/__support/FPUtil/FPBits.h:111:14: error: no matching function for call to 'bit_cast' 111 | : bits(cpp::bit_cast<UIntType>(x)) {} | ^~~~~~~~~~~~~~~~~~~~~~~ /usr/local/google/home/phosek/llvm/llvm-project/libc/src/__support/FPUtil/BasicOperations.h:22:13: note: in instantiation of function template specialization '__llvm_libc_18_0_0_git::fputil::FPBits<long double>::FPBits<long double, 0>' requested here 22 | FPBits<T> bits(x); | ^ /usr/local/google/home/phosek/llvm/llvm-project/libc/src/math/generic/fabsl.cpp:16:18: note: in instantiation of function template specialization '__llvm_libc_18_0_0_git::fputil::abs<long double, 0>' requested here 16 | return fputil::abs(x); | ^ /usr/local/google/home/phosek/llvm/llvm-project/libc/src/__support/CPP/bit.h:36:26: note: candidate template ignored: substitution failure [with To = UIntType, From = long double, $2 = cpp::enable_if_t<sizeof(UIntType) == sizeof(long double)>]: implicit instantiation of undefined template '__llvm_libc_18_0_0_git::cpp::enable_if<false>' 36 | LIBC_INLINE constexpr To bit_cast(const From &from) { | ^ 1 error generated. ```

[gchatelet]

On rv32, long double is IEEE754 binary 128 floating point number aka quadruple float.
The UIntType for it is UInt128 from:

llvm-project/libc/src/__support/UInt128.h

Lines 14 to 20 in dca432c

	#if !defined(__SIZEOF_INT128__)
	using UInt128 = LIBC_NAMESPACE::cpp::UInt<128>;
	using Int128 = LIBC_NAMESPACE::cpp::Int<128>;
	#else
	using UInt128 = __uint128_t;
	using Int128 = __int128_t;
	#endif

Because rv32 does not natively support __uint128_t this gets redirected to our software implementation via using UInt128 = LIBC_NAMESPACE::cpp::UInt<128>;

UInt is not trivially constructible because of zero initialized array here (the braces)

llvm-project/libc/src/__support/UInt.h

Line 33 in e0357e9

uint64_t val[WORDCOUNT]{};

The bit_cast implementation requires that the destination type is trivially constructible because the content of the object will be overridden - which may leave the object in a bogus state (resources may have been acquired during construction).

So I see two options here:

• We make UInt and Int behave like fundamental types (no zero initialization, default constructed objects are in an undefined state).
• We add another constructor in FPBit that kicks in when UIntType is UInt<N> and we explicitly initialize its internal val state using the memcpy semantics.

edit: We have the same issue here

llvm-project/libc/src/__support/FPUtil/x86_64/LongDoubleBits.h

Line 112 in 74e59e7

: bits(cpp::bit_cast<UIntType>(x)) {

[michaelrj-google]

I think making UInt trivially constructible makes the most sense, since that will make it better match the behavior of other integer types.

[gchatelet]

So unfortunately I think having UInt be bit_cast-able is a dead end.
For BigInt to be usable in constexpr expressions such as array initialization the val member needs to "mem-initialized".
Full GCC error:

/redacted/git/llvm-project/libc/src/math/generic/expm1.cpp: In function ‘__llvm_libc_18_0_0_git::Float128 __llvm_libc_18_0_0_git::poly_approx_f128(const Float128&)’:
/redacted/git/llvm-project/libc/src/math/generic/expm1.cpp:117:3: error: ‘constexpr __llvm_libc_18_0_0_git::cpp::BigInt<Bits, Signed>::BigInt(const uint64_t (&)[N]) [with long unsigned int N = 2; typename __llvm_libc_18_0_0_git::cpp::enable_if<(N <= WORDCOUNT), int>::type <anonymous> = 0; long unsigned int Bits = 128; bool Signed = false; uint64_t = long unsigned int]’ called in a constant expression
  117 |   };
      |   ^
In file included from /redacted/git/llvm-project/libc/src/__support/UInt128.h:12,
                 from /redacted/git/llvm-project/libc/src/__support/FPUtil/FloatProperties.h:12,
                 from /redacted/git/llvm-project/libc/src/__support/FPUtil/FPBits.h:17,
                 from /redacted/git/llvm-project/libc/src/math/generic/explogxf.h:17,
                 from /redacted/git/llvm-project/libc/src/math/generic/expm1.cpp:11:
/redacted/git/llvm-project/libc/src/__support/UInt.h:68:25: note: ‘constexpr __llvm_libc_18_0_0_git::cpp::BigInt<Bits, Signed>::BigInt(const uint64_t (&)[N]) [with long unsigned int N = 2; typename __llvm_libc_18_0_0_git::cpp::enable_if<(N <= WORDCOUNT), int>::type <anonymous> = 0; long unsigned int Bits = 128; bool Signed = false; uint64_t = long unsigned int]’ is not usable as a ‘constexpr’ function because:
   68 |   LIBC_INLINE constexpr BigInt(const uint64_t (&nums)[N]) {
      |                         ^~~~~~
/redacted/git/llvm-project/libc/src/__support/UInt.h:68:25: error: member ‘__llvm_libc_18_0_0_git::cpp::BigInt<128, false>::val’ must be initialized by mem-initializer in ‘constexpr’ constructor
/redacted/git/llvm-project/libc/src/__support/UInt.h:34:12: note: declared here
   34 |   uint64_t val[WORDCOUNT];
      |            ^~~

And if it's "mem-initialized" it is not trivially constructible anymore and so not bit_cast-able.
Clang is fine though but I think it's a legitimate error. To be honest, I also think that bit_casting for BigInt is dangerous as it requires careful control of its layout and blocks its evolution. I'll revert my patches and think about a better solution.

[lntue]

@gchatelet can you try to build the whole project or that part with C++20 standard for gcc to see if it fixes the problems?

[gchatelet]

@gchatelet can you try to build the whole project or that part with C++20 standard for gcc to see if it fixes the problems?

I just reverted it but let me try your suggestion. Are you referring to this part of the standard?

[lntue]

@gchatelet can you try to build the whole project or that part with C++20 standard for gcc to see if it fixes the problems?

I just reverted it but let me try your suggestion. Are you referring to this part of the standard?

Yes, I think that's the one blocking your attempt.

[gchatelet]

I'm running in a lot of errors when building with GCC-13 and C++20. I first thought it was because of poor compiler support but it seems it's well supported.
I didn't spot the one above although it might just be that compilation stops before reaching it.
Sample errors:

/usr/lib/gcc/x86_64-linux-gnu/13/include/mm_malloc.h: In function ‘void* _mm_malloc(size_t, size_t)’:
/usr/lib/gcc/x86_64-linux-gnu/13/include/mm_malloc.h:42:12: error: ‘malloc’ was not declared in this scope
   42 |     return malloc (__size);
      |            ^~~~~~
/usr/lib/gcc/x86_64-linux-gnu/13/include/mm_malloc.h:28:1: note: ‘malloc’ is defined in header ‘<cstdlib>’; did you forget to ‘#include <cstdlib>’?
   27 | #include <stdlib.h>
  +++ |+#include <cstdlib>
   28 | 
/usr/lib/gcc/x86_64-linux-gnu/13/include/mm_malloc.h: In function ‘void _mm_free(void*)’:
/usr/lib/gcc/x86_64-linux-gnu/13/include/mm_malloc.h:54:3: error: ‘free’ was not declared in this scope
   54 |   free (__ptr);
      |   ^~~~
/usr/lib/gcc/x86_64-linux-gnu/13/include/mm_malloc.h:54:3: note: ‘free’ is defined in header ‘<cstdlib>’; did you forget to ‘#include <cstdlib>’?

---

In file included from /redacted/git/llvm-project/libc/src/stdlib/atof.cpp:10:
/redacted/git/llvm-project/libc/src/stdlib/atof.cpp:16:28: error: ‘double __llvm_libc_18_0_0_git::atof(const char*)’ aliased to external symbol ‘atof’
   16 | LLVM_LIBC_FUNCTION(double, atof, (const char *str)) {
      |                            ^~~~
/redacted/git/llvm-project/libc/src/__support/common.h:28:34: note: in definition of macro ‘LLVM_LIBC_FUNCTION_IMPL’
   28 |   decltype(LIBC_NAMESPACE::name) name [[gnu::alias(#name)]];                   \
      |                                  ^~~~
/redacted/git/llvm-project/libc/src/stdlib/atof.cpp:16:1: note: in expansion of macro ‘LLVM_LIBC_FUNCTION’
   16 | LLVM_LIBC_FUNCTION(double, atof, (const char *str)) {
      | ^~~~~~~~~~~~~~~~~~
/redacted/git/llvm-project/libc/src/stdlib/atof.cpp:16:28: warning: ‘double __llvm_libc_18_0_0_git::atof(const char*)’ specifies less restrictive attributes than its target ‘double atof(const char*)’: ‘leaf’, ‘nonnull’, ‘nothrow’, ‘pure’ [-Wmissing-attributes]
   16 | LLVM_LIBC_FUNCTION(double, atof, (const char *str)) {
      |                            ^~~~
/redacted/git/llvm-project/libc/src/__support/common.h:28:34: note: in definition of macro ‘LLVM_LIBC_FUNCTION_IMPL’
   28 |   decltype(LIBC_NAMESPACE::name) name [[gnu::alias(#name)]];                   \
      |                                  ^~~~
/redacted/git/llvm-project/libc/src/stdlib/atof.cpp:16:1: note: in expansion of macro ‘LLVM_LIBC_FUNCTION’
   16 | LLVM_LIBC_FUNCTION(double, atof, (const char *str)) {
      | ^~~~~~~~~~~~~~~~~~
In file included from /usr/include/features.h:490,
                 from /usr/include/x86_64-linux-gnu/bits/libc-header-start.h:33,
                 from /usr/include/limits.h:26,
                 from /usr/lib/gcc/x86_64-linux-gnu/13/include/limits.h:205,
                 from /usr/lib/gcc/x86_64-linux-gnu/13/include/syslimits.h:7,
                 from /usr/lib/gcc/x86_64-linux-gnu/13/include/limits.h:34,
                 from /redacted/git/llvm-project/libc/src/__support/CPP/limits.h:16,
                 from /redacted/git/llvm-project/libc/src/__support/str_to_float.h:12,
                 from /redacted/git/llvm-project/libc/src/stdlib/atof.cpp:11:
/usr/include/x86_64-linux-gnu/bits/stdlib-float.h:25:1: note: ‘double __llvm_libc_18_0_0_git::atof(const char*)’ target declared here
   25 | __NTH (atof (const char *__nptr))
      | ^~~~~

[lntue]

Yes, I also see them when trying gcc with -Werror. Those are only warnings that you can skip for now. Probably we will need to clean up some of those extra attributes some time soon.