mirror of
https://codeberg.org/ziglang/zig.git
synced 2025-12-06 13:54:21 +00:00
compiler_rt: add subo
- approach by Hacker's Delight with wrapping subtraction - performance expected to be similar to addo - tests with all relevant combinations of min,max with -1,0,+1 and all combinations of sequences +-1,2,4..,max
This commit is contained in:
Jan Philipp Hafer
Andrew Kelley
parent
a15d2d582b
commit
fc59a04061
6 changed files with 289 additions and 1 deletions
|
|
@ -112,6 +112,12 @@ comptime {
|
|||
@export(__addodi4, .{ .name = "__addodi4", .linkage = linkage });
|
||||
const __addoti4 = @import("compiler_rt/addo.zig").__addoti4;
|
||||
@export(__addoti4, .{ .name = "__addoti4", .linkage = linkage });
|
||||
const __subosi4 = @import("compiler_rt/subo.zig").__subosi4;
|
||||
@export(__subosi4, .{ .name = "__subosi4", .linkage = linkage });
|
||||
const __subodi4 = @import("compiler_rt/subo.zig").__subodi4;
|
||||
@export(__subodi4, .{ .name = "__subodi4", .linkage = linkage });
|
||||
const __suboti4 = @import("compiler_rt/subo.zig").__suboti4;
|
||||
@export(__suboti4, .{ .name = "__suboti4", .linkage = linkage });
|
||||
const __mulosi4 = @import("compiler_rt/mulo.zig").__mulosi4;
|
||||
@export(__mulosi4, .{ .name = "__mulosi4", .linkage = linkage });
|
||||
const __mulodi4 = @import("compiler_rt/mulo.zig").__mulodi4;
|
||||
|
|
|
|||
|
|
@ -15,7 +15,7 @@ inline fn addoXi4_generic(comptime ST: type, a: ST, b: ST, overflow: *c_int) ST
|
|||
// and the sign of a+b+carry is the same as a (or equivalently b).
|
||||
// Slower routine: res = ~(a ^ b) & ((sum ^ a)
|
||||
// Faster routine: res = (sum ^ a) & (sum ^ b)
|
||||
// Oerflow occured, iff (res < 0)
|
||||
// Overflow occured, iff (res < 0)
|
||||
if (((sum ^ a) & (sum ^ b)) < 0)
|
||||
overflow.* = 1;
|
||||
return sum;
|
||||
|
|
|
|||
38
lib/std/special/compiler_rt/subo.zig
Normal file
38
lib/std/special/compiler_rt/subo.zig
Normal file
|
|
@ -0,0 +1,38 @@
|
|||
const builtin = @import("builtin");
|
||||
|
||||
// subo - subtract overflow
|
||||
// * return a-%b.
|
||||
// * return if a-b overflows => 1 else => 0
|
||||
// - suboXi4_generic as default
|
||||
|
||||
inline fn suboXi4_generic(comptime ST: type, a: ST, b: ST, overflow: *c_int) ST {
|
||||
@setRuntimeSafety(builtin.is_test);
|
||||
overflow.* = 0;
|
||||
var sum: ST = a -% b;
|
||||
// Hackers Delight: section Overflow Detection, subsection Signed Add/Subtract
|
||||
// Let sum = a -% b == a - b - carry == wraparound subtraction.
|
||||
// Overflow in a-b-carry occurs, iff a and b have opposite signs
|
||||
// and the sign of a-b-carry is opposite of a (or equivalently same as b).
|
||||
// Faster routine: res = (a ^ b) & (sum ^ a)
|
||||
// Slower routine: res = (sum^a) & ~(sum^b)
|
||||
// Overflow occured, iff (res < 0)
|
||||
if (((a ^ b) & (sum ^ a)) < 0)
|
||||
overflow.* = 1;
|
||||
return sum;
|
||||
}
|
||||
|
||||
pub fn __subosi4(a: i32, b: i32, overflow: *c_int) callconv(.C) i32 {
|
||||
return suboXi4_generic(i32, a, b, overflow);
|
||||
}
|
||||
pub fn __subodi4(a: i64, b: i64, overflow: *c_int) callconv(.C) i64 {
|
||||
return suboXi4_generic(i64, a, b, overflow);
|
||||
}
|
||||
pub fn __suboti4(a: i128, b: i128, overflow: *c_int) callconv(.C) i128 {
|
||||
return suboXi4_generic(i128, a, b, overflow);
|
||||
}
|
||||
|
||||
test {
|
||||
_ = @import("subosi4_test.zig");
|
||||
_ = @import("subodi4_test.zig");
|
||||
_ = @import("suboti4_test.zig");
|
||||
}
|
||||
81
lib/std/special/compiler_rt/subodi4_test.zig
Normal file
81
lib/std/special/compiler_rt/subodi4_test.zig
Normal file
|
|
@ -0,0 +1,81 @@
|
|||
const subo = @import("subo.zig");
|
||||
const std = @import("std");
|
||||
const testing = std.testing;
|
||||
const math = std.math;
|
||||
|
||||
fn test__subodi4(a: i64, b: i64) !void {
|
||||
var result_ov: c_int = undefined;
|
||||
var expected_ov: c_int = undefined;
|
||||
var result = subo.__subodi4(a, b, &result_ov);
|
||||
var expected: i64 = simple_subodi4(a, b, &expected_ov);
|
||||
try testing.expectEqual(expected, result);
|
||||
try testing.expectEqual(expected_ov, result_ov);
|
||||
}
|
||||
|
||||
// 2 cases on evaluating `a-b`:
|
||||
// 1. `a-b` may underflow, iff b>0 && a<0 and a-b < min <=> a<min+b
|
||||
// 2. `a-b` may overflow, iff b<0 && a>0 and a-b > max <=> a>max+b
|
||||
// `-b` evaluation may overflow, iff b==min, but this is handled by the hardware
|
||||
pub fn simple_subodi4(a: i64, b: i64, overflow: *c_int) i64 {
|
||||
overflow.* = 0;
|
||||
const min: i64 = math.minInt(i64);
|
||||
const max: i64 = math.maxInt(i64);
|
||||
if (((b > 0) and (a < min + b)) or
|
||||
((b < 0) and (a > max + b)))
|
||||
overflow.* = 1;
|
||||
return a -% b;
|
||||
}
|
||||
|
||||
test "subodi3" {
|
||||
const min: i64 = math.minInt(i64);
|
||||
const max: i64 = math.maxInt(i64);
|
||||
var i: i64 = 1;
|
||||
while (i < max) : (i *|= 2) {
|
||||
try test__subodi4(i, i);
|
||||
try test__subodi4(-i, -i);
|
||||
try test__subodi4(i, -i);
|
||||
try test__subodi4(-i, i);
|
||||
}
|
||||
|
||||
// edge cases
|
||||
// 0 - 0 = 0
|
||||
// MIN - MIN = 0
|
||||
// MAX - MAX = 0
|
||||
// 0 - MIN overflow
|
||||
// 0 - MAX = MIN+1
|
||||
// MIN - 0 = MIN
|
||||
// MAX - 0 = MAX
|
||||
// MIN - MAX overflow
|
||||
// MAX - MIN overflow
|
||||
try test__subodi4(0, 0);
|
||||
try test__subodi4(min, min);
|
||||
try test__subodi4(max, max);
|
||||
try test__subodi4(0, min);
|
||||
try test__subodi4(0, max);
|
||||
try test__subodi4(min, 0);
|
||||
try test__subodi4(max, 0);
|
||||
try test__subodi4(min, max);
|
||||
try test__subodi4(max, min);
|
||||
|
||||
// derived edge cases
|
||||
// MIN+1 - MIN = 1
|
||||
// MAX-1 - MAX = -1
|
||||
// 1 - MIN overflow
|
||||
// -1 - MIN = MAX
|
||||
// -1 - MAX = MIN
|
||||
// +1 - MAX = MIN+2
|
||||
// MIN - 1 overflow
|
||||
// MIN - -1 = MIN+1
|
||||
// MAX - 1 = MAX-1
|
||||
// MAX - -1 overflow
|
||||
try test__subodi4(min + 1, min);
|
||||
try test__subodi4(max - 1, max);
|
||||
try test__subodi4(1, min);
|
||||
try test__subodi4(-1, min);
|
||||
try test__subodi4(-1, max);
|
||||
try test__subodi4(1, max);
|
||||
try test__subodi4(min, 1);
|
||||
try test__subodi4(min, -1);
|
||||
try test__subodi4(max, -1);
|
||||
try test__subodi4(max, 1);
|
||||
}
|
||||
82
lib/std/special/compiler_rt/subosi4_test.zig
Normal file
82
lib/std/special/compiler_rt/subosi4_test.zig
Normal file
|
|
@ -0,0 +1,82 @@
|
|||
const subo = @import("subo.zig");
|
||||
const testing = @import("std").testing;
|
||||
|
||||
fn test__subosi4(a: i32, b: i32) !void {
|
||||
var result_ov: c_int = undefined;
|
||||
var expected_ov: c_int = undefined;
|
||||
var result = subo.__subosi4(a, b, &result_ov);
|
||||
var expected: i32 = simple_subosi4(a, b, &expected_ov);
|
||||
try testing.expectEqual(expected, result);
|
||||
try testing.expectEqual(expected_ov, result_ov);
|
||||
}
|
||||
|
||||
// 2 cases on evaluating `a-b`:
|
||||
// 1. `a-b` may underflow, iff b>0 && a<0 and a-b < min <=> a<min+b
|
||||
// 2. `a-b` may overflow, iff b<0 && a>0 and a-b > max <=> a>max+b
|
||||
// `-b` evaluation may overflow, iff b==min, but this is handled by the hardware
|
||||
pub fn simple_subosi4(a: i32, b: i32, overflow: *c_int) i32 {
|
||||
overflow.* = 0;
|
||||
const min: i32 = -2147483648;
|
||||
const max: i32 = 2147483647;
|
||||
if (((b > 0) and (a < min + b)) or
|
||||
((b < 0) and (a > max + b)))
|
||||
overflow.* = 1;
|
||||
return a -% b;
|
||||
}
|
||||
|
||||
test "subosi3" {
|
||||
// -2^31 <= i32 <= 2^31-1
|
||||
// 2^31 = 2147483648
|
||||
// 2^31-1 = 2147483647
|
||||
const min: i32 = -2147483648;
|
||||
const max: i32 = 2147483647;
|
||||
var i: i32 = 1;
|
||||
while (i < max) : (i *|= 2) {
|
||||
try test__subosi4(i, i);
|
||||
try test__subosi4(-i, -i);
|
||||
try test__subosi4(i, -i);
|
||||
try test__subosi4(-i, i);
|
||||
}
|
||||
|
||||
// edge cases
|
||||
// 0 - 0 = 0
|
||||
// MIN - MIN = 0
|
||||
// MAX - MAX = 0
|
||||
// 0 - MIN overflow
|
||||
// 0 - MAX = MIN+1
|
||||
// MIN - 0 = MIN
|
||||
// MAX - 0 = MAX
|
||||
// MIN - MAX overflow
|
||||
// MAX - MIN overflow
|
||||
try test__subosi4(0, 0);
|
||||
try test__subosi4(min, min);
|
||||
try test__subosi4(max, max);
|
||||
try test__subosi4(0, min);
|
||||
try test__subosi4(0, max);
|
||||
try test__subosi4(min, 0);
|
||||
try test__subosi4(max, 0);
|
||||
try test__subosi4(min, max);
|
||||
try test__subosi4(max, min);
|
||||
|
||||
// derived edge cases
|
||||
// MIN+1 - MIN = 1
|
||||
// MAX-1 - MAX = -1
|
||||
// 1 - MIN overflow
|
||||
// -1 - MIN = MAX
|
||||
// -1 - MAX = MIN
|
||||
// +1 - MAX = MIN+2
|
||||
// MIN - 1 overflow
|
||||
// MIN - -1 = MIN+1
|
||||
// MAX - 1 = MAX-1
|
||||
// MAX - -1 overflow
|
||||
try test__subosi4(min + 1, min);
|
||||
try test__subosi4(max - 1, max);
|
||||
try test__subosi4(1, min);
|
||||
try test__subosi4(-1, min);
|
||||
try test__subosi4(-1, max);
|
||||
try test__subosi4(1, max);
|
||||
try test__subosi4(min, 1);
|
||||
try test__subosi4(min, -1);
|
||||
try test__subosi4(max, -1);
|
||||
try test__subosi4(max, 1);
|
||||
}
|
||||
81
lib/std/special/compiler_rt/suboti4_test.zig
Normal file
81
lib/std/special/compiler_rt/suboti4_test.zig
Normal file
|
|
@ -0,0 +1,81 @@
|
|||
const subo = @import("subo.zig");
|
||||
const std = @import("std");
|
||||
const testing = std.testing;
|
||||
const math = std.math;
|
||||
|
||||
fn test__suboti4(a: i128, b: i128) !void {
|
||||
var result_ov: c_int = undefined;
|
||||
var expected_ov: c_int = undefined;
|
||||
var result = subo.__suboti4(a, b, &result_ov);
|
||||
var expected: i128 = simple_suboti4(a, b, &expected_ov);
|
||||
try testing.expectEqual(expected, result);
|
||||
try testing.expectEqual(expected_ov, result_ov);
|
||||
}
|
||||
|
||||
// 2 cases on evaluating `a-b`:
|
||||
// 1. `a-b` may underflow, iff b>0 && a<0 and a-b < min <=> a<min+b
|
||||
// 2. `a-b` may overflow, iff b<0 && a>0 and a-b > max <=> a>max+b
|
||||
// `-b` evaluation may overflow, iff b==min, but this is handled by the hardware
|
||||
pub fn simple_suboti4(a: i128, b: i128, overflow: *c_int) i128 {
|
||||
overflow.* = 0;
|
||||
const min: i128 = math.minInt(i128);
|
||||
const max: i128 = math.maxInt(i128);
|
||||
if (((b > 0) and (a < min + b)) or
|
||||
((b < 0) and (a > max + b)))
|
||||
overflow.* = 1;
|
||||
return a -% b;
|
||||
}
|
||||
|
||||
test "suboti3" {
|
||||
const min: i128 = math.minInt(i128);
|
||||
const max: i128 = math.maxInt(i128);
|
||||
var i: i128 = 1;
|
||||
while (i < max) : (i *|= 2) {
|
||||
try test__suboti4(i, i);
|
||||
try test__suboti4(-i, -i);
|
||||
try test__suboti4(i, -i);
|
||||
try test__suboti4(-i, i);
|
||||
}
|
||||
|
||||
// edge cases
|
||||
// 0 - 0 = 0
|
||||
// MIN - MIN = 0
|
||||
// MAX - MAX = 0
|
||||
// 0 - MIN overflow
|
||||
// 0 - MAX = MIN+1
|
||||
// MIN - 0 = MIN
|
||||
// MAX - 0 = MAX
|
||||
// MIN - MAX overflow
|
||||
// MAX - MIN overflow
|
||||
try test__suboti4(0, 0);
|
||||
try test__suboti4(min, min);
|
||||
try test__suboti4(max, max);
|
||||
try test__suboti4(0, min);
|
||||
try test__suboti4(0, max);
|
||||
try test__suboti4(min, 0);
|
||||
try test__suboti4(max, 0);
|
||||
try test__suboti4(min, max);
|
||||
try test__suboti4(max, min);
|
||||
|
||||
// derived edge cases
|
||||
// MIN+1 - MIN = 1
|
||||
// MAX-1 - MAX = -1
|
||||
// 1 - MIN overflow
|
||||
// -1 - MIN = MAX
|
||||
// -1 - MAX = MIN
|
||||
// +1 - MAX = MIN+2
|
||||
// MIN - 1 overflow
|
||||
// MIN - -1 = MIN+1
|
||||
// MAX - 1 = MAX-1
|
||||
// MAX - -1 overflow
|
||||
try test__suboti4(min + 1, min);
|
||||
try test__suboti4(max - 1, max);
|
||||
try test__suboti4(1, min);
|
||||
try test__suboti4(-1, min);
|
||||
try test__suboti4(-1, max);
|
||||
try test__suboti4(1, max);
|
||||
try test__suboti4(min, 1);
|
||||
try test__suboti4(min, -1);
|
||||
try test__suboti4(max, -1);
|
||||
try test__suboti4(max, 1);
|
||||
}
|
||||
Loading…
Add table
Reference in a new issue