stage1: add @shuffle() shufflevector support

I change the semantics of the mask operand, to make it a little more
flexible. There is no real danger in this because it is a compile-error
if you do it the LLVM way (and there is an appropiate error to tell you
this).

v2: avoid problems with double-free

2 files changed, 70 insertions, 0 deletions

diff --git a/test/compile_errors.zig b/test/compile_errors.zig
index 9d96d6f948..d9b4ee6a95 100644
--- a/test/compile_errors.zig
+++ b/test/compile_errors.zig
@@ -6485,6 +6485,19 @@ pub fn addCases(cases: *tests.CompileErrorContext) void {
     );
 
     cases.addTest(
+        "using LLVM syntax for @shuffle",
+        \\export fn entry() void {
+        \\    const v: @Vector(4, u32) = [4]u32{0, 1, 2, 3};
+        \\    const x: @Vector(4, u32) = [4]u32{4, 5, 6, 7};
+        \\    var z = @shuffle(u32, v, x, [8]i32{0, 1, 2, 3, 4, 5, 6, 7});
+        \\}
+    ,
+        "tmp.zig:4:39: error: mask index out of bounds",
+        "tmp.zig:4:39: note: when computing vector element at index 4",
+        "tmp.zig:4:39: note: selections from the second vector are specified with negative numbers",
+    );
+
+    cases.addTest(
         "nested vectors",
         \\export fn entry() void {
         \\    const V = @Vector(4, @Vector(4, u8));
diff --git a/test/stage1/behavior/shuffle.zig b/test/stage1/behavior/shuffle.zig
new file mode 100644
index 0000000000..70bff5991e
--- /dev/null
+++ b/test/stage1/behavior/shuffle.zig
@@ -0,0 +1,57 @@
+const std = @import("std");
+const mem = std.mem;
+const expect = std.testing.expect;
+
+test "@shuffle" {
+    const S = struct {
+        fn doTheTest() void {
+            var v: @Vector(4, i32) = [4]i32{ 2147483647, -2, 30, 40 };
+            var x: @Vector(4, i32) = [4]i32{ 1, 2147483647, 3, 4 };
+            const mask: @Vector(4, i32) = [4]i32{ 0, ~i32(2), 3, ~i32(3)};
+            var res = @shuffle(i32, v, x, mask);
+            expect(mem.eql(i32, ([4]i32)(res), [4]i32{ 2147483647, 3, 40, 4 }));
+
+            // Implicit cast from array (of mask)
+            res = @shuffle(i32, v, x, [4]i32{ 0, ~i32(2), 3, ~i32(3)});
+            expect(mem.eql(i32, ([4]i32)(res), [4]i32{ 2147483647, 3, 40, 4 }));
+
+            // Undefined
+            const mask2: @Vector(4, i32) = [4]i32{ 3, 1, 2, 0};
+            res = @shuffle(i32, v, undefined, mask2);
+            expect(mem.eql(i32, ([4]i32)(res), [4]i32{ 40, -2, 30, 2147483647}));
+
+            // Upcasting of b
+            var v2: @Vector(2, i32) = [2]i32{ 2147483647, undefined};
+            const mask3: @Vector(4, i32) = [4]i32{ ~i32(0), 2, ~i32(0), 3};
+            res = @shuffle(i32, x, v2, mask3);
+            expect(mem.eql(i32, ([4]i32)(res), [4]i32{ 2147483647, 3, 2147483647, 4 }));
+
+            // Upcasting of a
+            var v3: @Vector(2, i32) = [2]i32{ 2147483647, -2};
+            const mask4: @Vector(4, i32) = [4]i32{ 0, ~i32(2), 1, ~i32(3)};
+            res = @shuffle(i32, v3, x, mask4);
+            expect(mem.eql(i32, ([4]i32)(res), [4]i32{ 2147483647, 3, -2, 4 }));
+
+            // bool
+            {
+                var x2: @Vector(4, bool) = [4]bool{ false, true, false, true};
+                var v4: @Vector(2, bool) = [2]bool{ true, false};
+                const mask5: @Vector(4, i32) = [4]i32{ 0, ~i32(1), 1, 2};
+                var res2 = @shuffle(bool, x2, v4, mask5);
+                expect(mem.eql(bool, ([4]bool)(res2), [4]bool{ false, false, true, false }));
+            }
+
+            // FIXME re-enable when LLVM codegen is fixed
+            // https://bugs.llvm.org/show_bug.cgi?id=42803
+            if (false) {
+                var x2: @Vector(3, bool) = [3]bool{ false, true, false};
+                var v4: @Vector(2, bool) = [2]bool{ true, false};
+                const mask5: @Vector(4, i32) = [4]i32{ 0, ~i32(1), 1, 2};
+                var res2 = @shuffle(bool, x2, v4, mask5);
+                expect(mem.eql(bool, ([4]bool)(res2), [4]bool{ false, false, true, false }));
+            }
+        }
+    };
+    S.doTheTest();
+    comptime S.doTheTest();
+}