Incorrect upward rounding result of `fsub.s` instruction about riscv-isa-sim HOT 7 CLOSED

Computing the result then taking the ceil/floor does not have the same effect as setting the rounding mode, for multiple reasons. Anyway, this isn't a Spike bug.

from riscv-isa-sim.

Oh! I get it! Sorry, I misunderstand the underlying meaning of rounding mode.
In some implementations, the result will indeed be 0xfb418a35, which is an inaccurate result. But actually, 0xfb418a34 is more precise.

Thanks for your reply.

from riscv-isa-sim.

Intel produces 0xfb418a34, too:

$ cat test.c
#include <stdio.h>
#include <fenv.h>

int main()
{
  volatile int a = 0x284d52df;
  volatile int b = 0x7b418a35;
  volatile float f = *(volatile float*)&a;
  volatile float g = *(volatile float*)&b;

  fesetround(FE_UPWARD);
  volatile float h = f - g;

  printf("%x\n", *(volatile int*)&h);

  return 0;
}
$ gcc -O2 -lm test.c && ./a.out
fb418a34

from riscv-isa-sim.

It makes sense that, when rounding upward, the result would be 0xfb418a34 if the true result lies between 0xfb418a35 and 0xfb418a34, since the latter number is greater than the former.

from riscv-isa-sim.

$ cat main.cpp
#include <iostream>
#include <iomanip>
#include <cstdint>
#include <cstring>
#include <cmath>


// 函数声明
float hexToFloat(uint32_t hex);
uint32_t floatToHex(float value);

int main() {
    // 浮点数的十六进制表示
    uint32_t hex1 = 0x284d52df;
    uint32_t hex2 = 0x7b418a35;

    // 计算减法结果
    float result = hexToFloat(hex1) - hexToFloat(hex2);

    // 向上舍入
    if (result < 0) {
        result = ceilf(result);
    } else {
        result = floorf(result);
    }

    // 打印结果的十六进制表示
    uint32_t result_hex = floatToHex(result);
    std::cout << "最终结果的十六进制表示为: 0x" << std::hex << std::setw(8) << std::setfill('0') << result_hex << std::endl;

    return 0;
}

// 将十六进制表示的浮点数转换为浮点数
float hexToFloat(uint32_t hex) {
    float result;
    std::memcpy(&result, &hex, sizeof(float));
    return result;
}

// 将浮点数转换为十六进制表示并返回
uint32_t floatToHex(float value) {
    uint32_t result;
    std::memcpy(&result, &value, sizeof(uint32_t));
    return result;
}

$ g++ main.cpp -o main; ./main
最终结果的十六进制表示为: 0xfb418a35

Is there something wrong with main.cpp?

My processor model is Intel(R) Xeon(R) Platinum 8380 CPU @ 2.30GHz, Little Endian

from riscv-isa-sim.

You didn't set the rounding mode. The default rounding mode is round-to-nearest, ties-to-even. If you want to select round-up, you need to do so explicitly. (See my source code.)

from riscv-isa-sim.

Sorry, I have modified man.cpp(as the codes above) . But the result is still 0xfb418a35. I am trying your code.

from riscv-isa-sim.