From 0550198c9877c37c53b67f853968e3e4ce89c2ef Mon Sep 17 00:00:00 2001
From: Jan Philipp Hafer <jan.hafer@rwth-aachen.de>
Date: Mon, 13 Dec 2021 02:36:24 +0100
Subject: [PATCH] compiler_rt: simplify popcount "magic constants"

- magic constants are nicer to construct ie with
  (~@as(unsigned type, 0) / 3) == 0x55...55
- thanks to Stefan Kanthak for the idea
---
 lib/std/special/compiler_rt/popcount.zig | 55 ++++++++----------------
 1 file changed, 18 insertions(+), 37 deletions(-)

diff --git a/lib/std/special/compiler_rt/popcount.zig b/lib/std/special/compiler_rt/popcount.zig
index 2d4adec8c3676d1f677a30c5cf3450ca185a59c7..72513895db03c31bf4717387ebbfd778a3eb030e 100644
--- a/lib/std/special/compiler_rt/popcount.zig
+++ b/lib/std/special/compiler_rt/popcount.zig
@@ -9,45 +9,26 @@ const std = @import("std");
 //   SWAR-Masks and factors can be defined as 2-adic fractions
 // TAOCP: Combinational Algorithms, Bitwise Tricks And Techniques,
 //   subsubsection "Working with the rightmost bits" and "Sideways addition".
-fn popcountXi2_generic(comptime T: type) fn (a: T) callconv(.C) i32 {
+
+fn popcountXi2_generic(comptime ST: type) fn (a: ST) callconv(.C) i32 {
     return struct {
-        fn f(a: T) callconv(.C) i32 {
+        fn f(a: ST) callconv(.C) i32 {
             @setRuntimeSafety(builtin.is_test);
-
-            var x = switch (@bitSizeOf(T)) {
-                32 => @bitCast(u32, a),
-                64 => @bitCast(u64, a),
-                128 => @bitCast(u128, a),
-                else => unreachable,
-            };
-            const k1 = switch (@bitSizeOf(T)) { // -1/3
-                32 => @as(u32, 0x55555555),
-                64 => @as(u64, 0x55555555_55555555),
-                128 => @as(u128, 0x55555555_55555555_55555555_55555555),
-                else => unreachable,
-            };
-            const k2 = switch (@bitSizeOf(T)) { // -1/5
-                32 => @as(u32, 0x33333333),
-                64 => @as(u64, 0x33333333_33333333),
-                128 => @as(u128, 0x33333333_33333333_33333333_33333333),
-                else => unreachable,
-            };
-            const k4 = switch (@bitSizeOf(T)) { // -1/17
-                32 => @as(u32, 0x0f0f0f0f),
-                64 => @as(u64, 0x0f0f0f0f_0f0f0f0f),
-                128 => @as(u128, 0x0f0f0f0f_0f0f0f0f_0f0f0f0f_0f0f0f0f),
-                else => unreachable,
-            };
-            const kf = switch (@bitSizeOf(T)) { // -1/255
-                32 => @as(u32, 0x01010101),
-                64 => @as(u64, 0x01010101_01010101),
-                128 => @as(u128, 0x01010101_01010101_01010101_01010101),
-                else => unreachable,
-            };
-            x = x - ((x >> 1) & k1); // aggregate duos
-            x = (x & k2) + ((x >> 2) & k2); // aggregate nibbles
-            x = (x + (x >> 4)) & k4; // aggregate bytes
-            x = (x *% kf) >> @bitSizeOf(T) - 8; // 8 most significant bits of x + (x<<8) + (x<<16) + ..
+            const UT = switch (ST) {
+                i32 => u32,
+                i64 => u64,
+                i128 => u128,
+                else => unreachable,
+            };
+            var x = @bitCast(UT, a);
+            x -= (x >> 1) & (~@as(UT, 0) / 3); // 0x55...55, aggregate duos
+            x = ((x >> 2) & (~@as(UT, 0) / 5)) // 0x33...33, aggregate nibbles
+            + (x & (~@as(UT, 0) / 5));
+            x += x >> 4;
+            x &= ~@as(UT, 0) / 17; // 0x0F...0F, aggregate bytes
+            // 8 most significant bits of x + (x<<8) + (x<<16) + ..
+            x *%= ~@as(UT, 0) / 255; // 0x01...01
+            x >>= (@bitSizeOf(ST) - 8);
             return @intCast(i32, x);
         }
     }.f;
-- 
2.54.0