Commit 0512be227c

LemonBoy <thatlemon@gmail.com>
2020-03-27 20:10:59
compiler-rt: Fix __floatunditf
This builtin converts a u64 into a f128, not a u128 into a f128. Fixes some weird-ass crashes that happened only on AArch64 systems.
1 parent aec4967
Changed files (2)
lib
std
lib/std/special/compiler_rt/floatunditf.zig
@@ -2,7 +2,7 @@ const builtin = @import("builtin");
 const is_test = builtin.is_test;
 const std = @import("std");
 
-pub fn __floatunditf(a: u128) callconv(.C) f128 {
+pub fn __floatunditf(a: u64) callconv(.C) f128 {
     @setRuntimeSafety(is_test);
 
     if (a == 0) {
@@ -14,11 +14,11 @@ pub fn __floatunditf(a: u128) callconv(.C) f128 {
     const exponent_bias = (1 << (exponent_bits - 1)) - 1;
     const implicit_bit = 1 << mantissa_bits;
 
-    const exp = (u128.bit_count - 1) - @clz(u128, a);
-    const shift = mantissa_bits - @intCast(u7, exp);
+    const exp: u128 = (u64.bit_count - 1) - @clz(u64, a);
+    const shift: u7 = mantissa_bits - @intCast(u7, exp);
 
-    var result: u128 align(16) = (a << shift) ^ implicit_bit;
-    result += (@intCast(u128, exp) + exponent_bias) << mantissa_bits;
+    var result: u128 = (@intCast(u128, a) << shift) ^ implicit_bit;
+    result += (exp + exponent_bias) << mantissa_bits;
 
     return @bitCast(f128, result);
 }
lib/std/special/compiler_rt/floatunditf_test.zig
@@ -1,6 +1,6 @@
 const __floatunditf = @import("floatunditf.zig").__floatunditf;
 
-fn test__floatunditf(a: u128, expected_hi: u64, expected_lo: u64) void {
+fn test__floatunditf(a: u64, expected_hi: u64, expected_lo: u64) void {
     const x = __floatunditf(a);
 
     const x_repr = @bitCast(u128, x);