add popcountdi2 to compiler_rt

CMakeLists.txt

@@ -655,6 +655,7 @@ set(ZIG_STD_FILES
     "special/compiler_rt/floatuntitf.zig"
     "special/compiler_rt/muloti4.zig"
     "special/compiler_rt/multi3.zig"
+    "special/compiler_rt/popcountdi2.zig"
     "special/compiler_rt/truncXfYf2.zig"
     "special/compiler_rt/udivmod.zig"
     "special/compiler_rt/udivmoddi4.zig"

std/special/compiler_rt.zig

@@ -67,6 +67,7 @@ comptime {
     @export("__fixtfti", @import("compiler_rt/fixtfti.zig").__fixtfti, linkage);
 
     @export("__udivmoddi4", @import("compiler_rt/udivmoddi4.zig").__udivmoddi4, linkage);
+    @export("__popcountdi2", @import("compiler_rt/popcountdi2.zig").__popcountdi2, linkage);
 
     @export("__udivsi3", __udivsi3, linkage);
     @export("__udivdi3", __udivdi3, linkage);

std/special/compiler_rt/popcountdi2.zig

@@ -0,0 +1,24 @@
+const builtin = @import("builtin");
+const compiler_rt = @import("../compiler_rt.zig");
+
+// ported from llvm compiler-rt 8.0.0rc3 95e1c294cb0415a377a7b1d6c7c7d4f89e1c04e4
+pub extern fn __popcountdi2(a: i64) i32 {
+    var x2 = @bitCast(u64, a);
+    x2 = x2 - ((x2 >> 1) & 0x5555555555555555);
+    // Every 2 bits holds the sum of every pair of bits (32)
+    x2 = ((x2 >> 2) & 0x3333333333333333) + (x2 & 0x3333333333333333);
+    // Every 4 bits holds the sum of every 4-set of bits (3 significant bits) (16)
+    x2 = (x2 + (x2 >> 4)) & 0x0F0F0F0F0F0F0F0F;
+    // Every 8 bits holds the sum of every 8-set of bits (4 significant bits) (8)
+    var x: u32 = @truncate(u32, x2 + (x2 >> 32));
+    // The lower 32 bits hold four 16 bit sums (5 significant bits).
+    //   Upper 32 bits are garbage */
+    x = x + (x >> 16);
+    // The lower 16 bits hold two 32 bit sums (6 significant bits).
+    //   Upper 16 bits are garbage */
+    return @bitCast(i32, (x + (x >> 8)) & 0x0000007F); // (7 significant bits)
+}
+
+test "import popcountdi2" {
+    _ = @import("popcountdi2_test.zig");
+}

std/special/compiler_rt/popcountdi2_test.zig

@@ -0,0 +1,27 @@
+const __popcountdi2 = @import("popcountdi2.zig").__popcountdi2;
+const testing = @import("std").testing;
+
+fn naive_popcount(a_param: i64) i32 {
+    var a = a_param;
+    var r: i32 = 0;
+    while (a != 0) : (a = @bitCast(i64, @bitCast(u64, a) >> 1)) {
+        r += @intCast(i32, a & 1);
+    }
+    return r;
+}
+
+fn test__popcountdi2(a: i64) void {
+    const x = __popcountdi2(a);
+    const expected = naive_popcount(a);
+    testing.expect(expected == x);
+}
+
+test "popcountdi2" {
+    test__popcountdi2(0);
+    test__popcountdi2(1);
+    test__popcountdi2(2);
+    test__popcountdi2(@bitCast(i64, u64(0xFFFFFFFFFFFFFFFD)));
+    test__popcountdi2(@bitCast(i64, u64(0xFFFFFFFFFFFFFFFE)));
+    test__popcountdi2(@bitCast(i64, u64(0xFFFFFFFFFFFFFFFF)));
+    // TODO some fuzz testing
+}