大学习nan

近日观测到dram和冰冰对RISC-V isa manual float point部分进行了补充说明的patch。

冰冰解释了 printf 打印 float 会丢失符号位，因为有 float 转 double 的过程，但是直接printf double就没有问题。

icenowy@edelgard [ tmp ] $ gcc test.c 
icenowy@edelgard [ tmp ] $ ./a.out 
-nan
icenowy@edelgard [ tmp ] $ cat test.c 
#include <stdio.h>

union u {
        unsigned long i;
        double f;
};

int main()
{
        union u u;

        u.i = 0xfff0000001000000ul;

        printf("%f\n", u.f);

        return 0;
}

恰好我需要修的llvm mlir测例中出现了如下内容：

  // CHECK-NEXT: -nan
  %h = arith.constant 1.0 : f64
  %h_p = arith.constant 0xfff0000001000000 : f64
  call @func_powff64(%h, %h_p) : (f64, f64) -> ()

报错：

# .---command stderr------------
# | /home/chenyixuan/llvm-project/mlir/test/mlir-runner/test-expand-math-approx.mlir:236:17: error: CHECK-NEXT: expected string not found in input
# |  // CHECK-NEXT: -nan
# |                 ^
# | <stdin>:39:4: note: scanning from here
# | inf
# |    ^
# | <stdin>:40:1: note: possible intended match here
# | nan
# | ^
# | 
# | Input file: <stdin>
# | Check file: /home/chenyixuan/llvm-project/mlir/test/mlir-runner/test-expand-math-approx.mlir
# | 
# | -dump-input=help explains the following input dump.
# | 
# | Input was:
# | <<<<<<
# |             .
# |             .
# |             .
# |            34: 2.343 
# |            35: 0.176171 
# |            36: 1 
# |            37: 6.62637 
# |            38: nan 
# |            39: inf 
# | next:236'0        X error: no match found
# |            40: nan 
# | next:236'0     ~~~~
# | next:236'1     ?    possible intended match
# |            41: nan 
# | next:236'0     ~~~~
# |            42: 1 
# | next:236'0     ~~
# |            43: 4 
# | next:236'0     ~~
# |            44: 0.25 
# | next:236'0     ~~~~~
# |            45: 0 
# | next:236'0     ~~
# |             .
# |             .
# |             .
# | >>>>>>
# `-----------------------------
# error: command failed with exit status: 1

于是强大冰冰帮我复现了fail：

(ssh)root@lpi4a66 [ tmp ] # gcc test.c -lm -fno-builtin
(ssh)root@lpi4a66 [ tmp ] # ./a.out 
nan
(ssh)root@lpi4a66 [ tmp ] # cat test.c 
#include <stdio.h>
#include <math.h>

union u {
        unsigned long i;
        double f;
};

int main()
{
        union u u;

        u.i = 0xfff0000001000000ul;

        printf("%f\n", pow(1.0, u.f));

        return 0;
}

RISC-V 的 nan 在运算之后也会丢失符号位，于是我放宽了测例对符号位的限制。 // CHECK-NEXT: {{-?}}nan