我需要同时转换 32 位和 64 位 无符号整数 转换为 xmm 寄存器中的浮点值。有 x86 指令可以将有符号整数转换为单精度和 double 浮点值，但没有用于无符号整数。
奖励:如何将 xmm 寄存器中的浮点值转换为 32 位和 64 位无符号整数？
最佳答案
这是 GCC 生成的内容。我将它们包装在函数中，但您可以轻松删除堆栈处理。不是所有的都使用SSE来做实际工作(ulonglong转换没有)，如果你找到相应的说明，请告诉我。 Clang 生成的几乎相同。
% cat tofloats.c
double ulonglong2double(unsigned long long a) {
return a;
}
float ulonglong2float(unsigned long long a) {
return a;
}
double uint2double(unsigned int a) {
return a;
}
float uint2float(unsigned int a) {
return a;
}
% gcc -msse4.2 -g -Os -c tofloats.c && objdump -d tofloats.o
00000000 <ulonglong2double>:
0: 55 push %ebp
1: 89 e5 mov %esp,%ebp
3: 83 ec 10 sub $0x10,%esp
6: 8b 55 0c mov 0xc(%ebp),%edx
9: 8b 45 08 mov 0x8(%ebp),%eax
c: 89 55 f4 mov %edx,-0xc(%ebp)
f: 85 d2 test %edx,%edx
11: 89 45 f0 mov %eax,-0x10(%ebp)
14: df 6d f0 fildll -0x10(%ebp)
17: 79 06 jns 1f <ulonglong2double+0x1f>
19: d8 05 00 00 00 00 fadds 0x0
1f: dd 5d f8 fstpl -0x8(%ebp)
22: dd 45 f8 fldl -0x8(%ebp)
25: c9 leave
26: c3 ret
00000027 <ulonglong2float>:
27: 55 push %ebp
28: 89 e5 mov %esp,%ebp
2a: 83 ec 10 sub $0x10,%esp
2d: 8b 55 0c mov 0xc(%ebp),%edx
30: 8b 45 08 mov 0x8(%ebp),%eax
33: 89 55 f4 mov %edx,-0xc(%ebp)
36: 85 d2 test %edx,%edx
38: 89 45 f0 mov %eax,-0x10(%ebp)
3b: df 6d f0 fildll -0x10(%ebp)
3e: 79 06 jns 46 <ulonglong2float+0x1f>
40: d8 05 00 00 00 00 fadds 0x0
46: d9 5d fc fstps -0x4(%ebp)
49: d9 45 fc flds -0x4(%ebp)
4c: c9 leave
4d: c3 ret
0000004e <uint2double>:
4e: 55 push %ebp
4f: 89 e5 mov %esp,%ebp
51: 83 ec 08 sub $0x8,%esp
54: 66 0f 6e 45 08 movd 0x8(%ebp),%xmm0
59: 66 0f d6 45 f8 movq %xmm0,-0x8(%ebp)
5e: df 6d f8 fildll -0x8(%ebp)
61: c9 leave
62: c3 ret
00000063 <uint2float>:
63: 55 push %ebp
64: 89 e5 mov %esp,%ebp
66: 83 ec 08 sub $0x8,%esp
69: 66 0f 6e 45 08 movd 0x8(%ebp),%xmm0
6e: 66 0f d6 45 f8 movq %xmm0,-0x8(%ebp)
73: df 6d f8 fildll -0x8(%ebp)
76: c9 leave
77: c3 ret
% cat toints.c
unsigned long long float2ulonglong(float a) {
return a;
}
unsigned long long double2ulonglong(double a) {
return a;
}
unsigned int float2uint(float a) {
return a;
}
unsigned int double2uint(double a) {
return a;
}
% gcc -msse4.2 -g -Os -c toints.c && objdump -d toints.o
00000000 <float2ulonglong>:
0: 55 push %ebp
1: 89 e5 mov %esp,%ebp
3: 53 push %ebx
4: 83 ec 0c sub $0xc,%esp
7: d9 45 08 flds 0x8(%ebp)
a: d9 05 00 00 00 00 flds 0x0
10: d9 c9 fxch %st(1)
12: db e9 fucomi %st(1),%st
14: 73 0d jae 23 <float2ulonglong+0x23>
16: dd d9 fstp %st(1)
18: dd 4d f0 fisttpll -0x10(%ebp)
1b: 8b 45 f0 mov -0x10(%ebp),%eax
1e: 8b 55 f4 mov -0xc(%ebp),%edx
21: eb 13 jmp 36 <float2ulonglong+0x36>
23: de e1 fsubp %st,%st(1)
25: dd 4d f0 fisttpll -0x10(%ebp)
28: 8b 55 f4 mov -0xc(%ebp),%edx
2b: 8b 45 f0 mov -0x10(%ebp),%eax
2e: 8d 8a 00 00 00 80 lea -0x80000000(%edx),%ecx
34: 89 ca mov %ecx,%edx
36: 83 c4 0c add $0xc,%esp
39: 5b pop %ebx
3a: 5d pop %ebp
3b: c3 ret
0000003c <double2ulonglong>:
3c: 55 push %ebp
3d: 89 e5 mov %esp,%ebp
3f: 53 push %ebx
40: 83 ec 0c sub $0xc,%esp
43: dd 45 08 fldl 0x8(%ebp)
46: d9 05 00 00 00 00 flds 0x0
4c: d9 c9 fxch %st(1)
4e: db e9 fucomi %st(1),%st
50: 73 0d jae 5f <double2ulonglong+0x23>
52: dd d9 fstp %st(1)
54: dd 4d f0 fisttpll -0x10(%ebp)
57: 8b 45 f0 mov -0x10(%ebp),%eax
5a: 8b 55 f4 mov -0xc(%ebp),%edx
5d: eb 13 jmp 72 <double2ulonglong+0x36>
5f: de e1 fsubp %st,%st(1)
61: dd 4d f0 fisttpll -0x10(%ebp)
64: 8b 55 f4 mov -0xc(%ebp),%edx
67: 8b 45 f0 mov -0x10(%ebp),%eax
6a: 8d 8a 00 00 00 80 lea -0x80000000(%edx),%ecx
70: 89 ca mov %ecx,%edx
72: 83 c4 0c add $0xc,%esp
75: 5b pop %ebx
76: 5d pop %ebp
77: c3 ret
00000078 <float2uint>:
78: 55 push %ebp
79: 89 e5 mov %esp,%ebp
7b: 83 ec 08 sub $0x8,%esp
7e: d9 45 08 flds 0x8(%ebp)
81: dd 4d f8 fisttpll -0x8(%ebp)
84: 8b 45 f8 mov -0x8(%ebp),%eax
87: c9 leave
88: c3 ret
00000089 <double2uint>:
89: 55 push %ebp
8a: 89 e5 mov %esp,%ebp
8c: 83 ec 08 sub $0x8,%esp
8f: dd 45 08 fldl 0x8(%ebp)
92: dd 4d f8 fisttpll -0x8(%ebp)
95: 8b 45 f8 mov -0x8(%ebp),%eax
98: c9 leave
99: c3 ret
fisttpll:存储带有截断的整数
FISTTP converts the value in ST into a signed integer using truncation (chop) as rounding mode, transfers the result to the destination, and pop ST. FISTTP accepts word, short integer, and long integer destinations.
fucomi:比较浮点值并设置 EFLAGS
Performs an unordered comparison of the contents of registers ST(0) and ST(i) and sets the status flags ZF, PF, and CF in the EFLAGS register according to the results (see the table below). The sign of zero is ignored for comparisons, so that –0.0 is equal to +0.0.
关于assembly - 如何在 x86(32 位)程序集中将无符号整数转换为 float ？，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/11406654/