Bug 10905 – [ICE](ctfeexpr.c line 355) with ulong2 in structs

import core.simd: ulong2; struct Foo { enum ulong2 y = 1; } struct Bar { ulong2 x; bool spam() const { return x == Foo.y; } } void main() {} DMD 2.064alpha gives: test.d(3): Error: Internal Compiler Error: CTFE literal cast(__vector(ulong[2u]))1LU Assertion failure: '0' on line 355 in file 'ctfeexpr.c'

Why does this say x86 on Windows for the platform? x86 on Windows doesn't support SIMD at all.

Commits pushed to master at https://github.com/D-Programming-Language/dmd https://github.com/D-Programming-Language/dmd/commit/74dbc6e48b5e3da2944438804d5cdd055f6fb129 Issue 10905 - [CTFE] Fix [ICE](ctfeexpr.c line 355) with ulong2 in structs https://github.com/D-Programming-Language/dmd/commit/99ff22023fcaf7de49a71bd2e1691523943b8fd5 Merge pull request #2937 from ibuclaw/issue10905 Issue 10905 - [CTFE] Fix [ICE](ctfeexpr.c line 355) with ulong2 in structs

I guess x86 Windows is bearophile's platform of choice, and that's the compiler he used when he found the bug.

The ICE is fixex, but this code gives: import core.simd: ulong2; void main() { ulong2 x; immutable ulong2 y; bool b = x == y; } temp.d(5): Error: incompatible types for ((cast(const(__vector(ulong[2])))x) == (cast(const(__vector(ulong[2])))y)): 'const(__vector(ulong[2]))' and 'const(__vector(ulong[2]))'

See the testcase in the pull request. Currently the frontend explicitly disallows comparing vectors using == expression. This seems like an odd limitation to me, but I'm not arguing it. Please raise a new pull request if you wish to change the language behaviour. Regards Iain

Err... I mean please raise a new bug report.

One thing to note is that the result of a vector == comparison would not be a boolean. import core.simd; int4 a = [1,2,3,4]; int4 b = [3,2,1,4]; if (a == b) { } // Error: vector comparison returning a boolean. int4 c = (a == b); assert(c.array = [0, -1, 0, -1]);

(In reply to comment #9) > One thing to note is that the result of a vector == comparison would not be a > boolean. > > import core.simd; > > int4 a = [1,2,3,4]; > int4 b = [3,2,1,4]; > > if (a == b) { } // Error: vector comparison returning a boolean. > > int4 c = (a == b); > assert(c.array = [0, -1, 0, -1]); That seems acceptable. But Manu is much more expert than me on such matters, so we should listen to his opinion on this. A possible semantics for: if (a == b) {} else {} Is to run the first branch if all items are equal ([-1, -1, -1, -1]) and the second branch otherwise. I am not sure but I think some kind of Intel compiler works like this.