| Issue |
60625
|
| Summary |
c++: converting uint32_t xvalue to rvalue reference to float does not work in this:
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
imzhuli
|
Please take a look at the following code, which tends to break float into uint parts for transfer, (not for calculation, so endian is not considered) , and to build it back.
The MergeFloat_Wrongly does not generate expected result as MergeFloat does:
inline void SplitFloat32(bool &sign_bit, uint &exponent, uint &mantissa, const float value)
{
float copy = value;
uint32_t *p = (uint32_t *)©
sign_bit = (*p & 0x80000000) != 0;
exponent = (*p & 0x7F800000) >> 23;
mantissa = (*p & 0x007FFFFF);
}
float MergeFloat(const bool sign_bit, const uint exponent, const uint mantissa)
{
// this works:
uint32_t Temp = uint32_t((sign_bit ? 0x80000000 : 0) | (exponent << 23) | mantissa);
return (float&)Temp;
}
float MergeFloat_Wrongly(const bool sign_bit, const uint exponent, const uint mantissa)
{
// THIS DOES NOT WORK, the compiler directly convert uint32 xvalue to float, which is expected to be loaded directly
auto f = (float&&)(uint32_t((sign_bit ? 0x80000000 : 0) | (exponent << 23) | mantissa));
return f;
}
///////// THE ASM IN MEMORY ////////
0x100003080 <+0>: pushq %rbp
0x100003081 <+1>: movq %rsp, %rbp
0x100003084 <+4>: movb %dil, %al
0x100003087 <+7>: andb $0x1, %al
0x100003089 <+9>: movb %al, -0x1(%rbp)
0x10000308c <+12>: movl %esi, -0x8(%rbp)
0x10000308f <+15>: movl %edx, -0xc(%rbp)
0x100003092 <+18>: movzbl -0x1(%rbp), %eax
0x100003096 <+22>: shll $0x1f, %eax
0x100003099 <+25>: movl -0x8(%rbp), %ecx
0x10000309c <+28>: shll $0x17, %ecx
0x10000309f <+31>: orl %ecx, %eax
0x1000030a1 <+33>: movl -0xc(%rbp), %ecx
0x1000030a4 <+36>: orl %ecx, %eax
-> 0x1000030a6 <+38>: cvtsi2ss %rax, %xmm0
0x1000030ab <+43>: movss %xmm0, -0x14(%rbp)
it seems it did a normal direct conversion from uint32(eax)->64(rax)->double(xmm0) with cvtsi2ss
Environment:
Apple clang version 14.0.0 (clang-1400.0.29.202)
Target: x86_64-apple-darwin22.2.0
(btw: g++ seems to have same issue)
I hope it's my understanding of c++ grammar that went wrong.
Anyway, thanks for any possible reply
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
