zig/lib/compiler_rt/tan.zig
Koki Ueha 878b7b80c1 libc: Prevent FCSEL instruction from being used to avoid raising an unintended exception
If you write an if expression in mem.doNotOptimizeAway like
doNotOptimizeAway(if (ix < 0x00100000) x / 0x1p120 else x + 0x1p120);,
FCSEL instruction is used on AArch64.
FCSEL instruction selects one of the two registers according to
the condition and copies its value.
In this example, `x / 0x1p120` and `x + 0x1p120` are expressions
that raise different floating-point exceptions.
However, since both are actually evaluated before the FCSEL
instruction, the exception not intended by the programmer may
also be raised.

To prevent FCSEL instruction from being used here, this commit
splits doNotOptimizeAway in two.
2025-06-15 04:01:43 -04:00

183 lines
6.0 KiB
Zig

//! Ported from musl, which is licensed under the MIT license:
//! https://git.musl-libc.org/cgit/musl/tree/COPYRIGHT
//!
//! https://git.musl-libc.org/cgit/musl/tree/src/math/tanf.c
//! https://git.musl-libc.org/cgit/musl/tree/src/math/tan.c
//! https://golang.org/src/math/tan.go
const std = @import("std");
const builtin = @import("builtin");
const math = std.math;
const mem = std.mem;
const expect = std.testing.expect;
const kernel = @import("trig.zig");
const rem_pio2 = @import("rem_pio2.zig").rem_pio2;
const rem_pio2f = @import("rem_pio2f.zig").rem_pio2f;
const arch = builtin.cpu.arch;
const common = @import("common.zig");
pub const panic = common.panic;
comptime {
@export(&__tanh, .{ .name = "__tanh", .linkage = common.linkage, .visibility = common.visibility });
@export(&tanf, .{ .name = "tanf", .linkage = common.linkage, .visibility = common.visibility });
@export(&tan, .{ .name = "tan", .linkage = common.linkage, .visibility = common.visibility });
@export(&__tanx, .{ .name = "__tanx", .linkage = common.linkage, .visibility = common.visibility });
if (common.want_ppc_abi) {
@export(&tanq, .{ .name = "tanf128", .linkage = common.linkage, .visibility = common.visibility });
}
@export(&tanq, .{ .name = "tanq", .linkage = common.linkage, .visibility = common.visibility });
@export(&tanl, .{ .name = "tanl", .linkage = common.linkage, .visibility = common.visibility });
}
pub fn __tanh(x: f16) callconv(.c) f16 {
// TODO: more efficient implementation
return @floatCast(tanf(x));
}
pub fn tanf(x: f32) callconv(.c) f32 {
// Small multiples of pi/2 rounded to double precision.
const t1pio2: f64 = 1.0 * math.pi / 2.0; // 0x3FF921FB, 0x54442D18
const t2pio2: f64 = 2.0 * math.pi / 2.0; // 0x400921FB, 0x54442D18
const t3pio2: f64 = 3.0 * math.pi / 2.0; // 0x4012D97C, 0x7F3321D2
const t4pio2: f64 = 4.0 * math.pi / 2.0; // 0x401921FB, 0x54442D18
var ix: u32 = @bitCast(x);
const sign = ix >> 31 != 0;
ix &= 0x7fffffff;
if (ix <= 0x3f490fda) { // |x| ~<= pi/4
if (ix < 0x39800000) { // |x| < 2**-12
// raise inexact if x!=0 and underflow if subnormal
if (common.want_float_exceptions) {
if (ix < 0x00800000) {
mem.doNotOptimizeAway(x / 0x1p120);
} else {
mem.doNotOptimizeAway(x + 0x1p120);
}
}
return x;
}
return kernel.__tandf(x, false);
}
if (ix <= 0x407b53d1) { // |x| ~<= 5*pi/4
if (ix <= 0x4016cbe3) { // |x| ~<= 3pi/4
return kernel.__tandf((if (sign) x + t1pio2 else x - t1pio2), true);
} else {
return kernel.__tandf((if (sign) x + t2pio2 else x - t2pio2), false);
}
}
if (ix <= 0x40e231d5) { // |x| ~<= 9*pi/4
if (ix <= 0x40afeddf) { // |x| ~<= 7*pi/4
return kernel.__tandf((if (sign) x + t3pio2 else x - t3pio2), true);
} else {
return kernel.__tandf((if (sign) x + t4pio2 else x - t4pio2), false);
}
}
// tan(Inf or NaN) is NaN
if (ix >= 0x7f800000) {
return x - x;
}
var y: f64 = undefined;
const n = rem_pio2f(x, &y);
return kernel.__tandf(y, n & 1 != 0);
}
pub fn tan(x: f64) callconv(.c) f64 {
var ix = @as(u64, @bitCast(x)) >> 32;
ix &= 0x7fffffff;
// |x| ~< pi/4
if (ix <= 0x3fe921fb) {
if (ix < 0x3e400000) { // |x| < 2**-27
// raise inexact if x!=0 and underflow if subnormal
if (common.want_float_exceptions) {
if (ix < 0x00100000) {
mem.doNotOptimizeAway(x / 0x1p120);
} else {
mem.doNotOptimizeAway(x + 0x1p120);
}
}
return x;
}
return kernel.__tan(x, 0.0, false);
}
// tan(Inf or NaN) is NaN
if (ix >= 0x7ff00000) {
return x - x;
}
var y: [2]f64 = undefined;
const n = rem_pio2(x, &y);
return kernel.__tan(y[0], y[1], n & 1 != 0);
}
pub fn __tanx(x: f80) callconv(.c) f80 {
// TODO: more efficient implementation
return @floatCast(tanq(x));
}
pub fn tanq(x: f128) callconv(.c) f128 {
// TODO: more correct implementation
return tan(@floatCast(x));
}
pub fn tanl(x: c_longdouble) callconv(.c) c_longdouble {
switch (@typeInfo(c_longdouble).float.bits) {
16 => return __tanh(x),
32 => return tanf(x),
64 => return tan(x),
80 => return __tanx(x),
128 => return tanq(x),
else => @compileError("unreachable"),
}
}
test "tan" {
try expect(tan(@as(f32, 0.0)) == tanf(0.0));
try expect(tan(@as(f64, 0.0)) == tan(0.0));
}
test "tan32" {
const epsilon = 0.00001;
try expect(math.approxEqAbs(f32, tanf(0.0), 0.0, epsilon));
try expect(math.approxEqAbs(f32, tanf(0.2), 0.202710, epsilon));
try expect(math.approxEqAbs(f32, tanf(0.8923), 1.240422, epsilon));
try expect(math.approxEqAbs(f32, tanf(1.5), 14.101420, epsilon));
try expect(math.approxEqAbs(f32, tanf(37.45), -0.254397, epsilon));
try expect(math.approxEqAbs(f32, tanf(89.123), 2.285852, epsilon));
}
test "tan64" {
const epsilon = 0.000001;
try expect(math.approxEqAbs(f64, tan(0.0), 0.0, epsilon));
try expect(math.approxEqAbs(f64, tan(0.2), 0.202710, epsilon));
try expect(math.approxEqAbs(f64, tan(0.8923), 1.240422, epsilon));
try expect(math.approxEqAbs(f64, tan(1.5), 14.101420, epsilon));
try expect(math.approxEqAbs(f64, tan(37.45), -0.254397, epsilon));
try expect(math.approxEqAbs(f64, tan(89.123), 2.2858376, epsilon));
}
test "tan32.special" {
try expect(tanf(0.0) == 0.0);
try expect(tanf(-0.0) == -0.0);
try expect(math.isNan(tanf(math.inf(f32))));
try expect(math.isNan(tanf(-math.inf(f32))));
try expect(math.isNan(tanf(math.nan(f32))));
}
test "tan64.special" {
try expect(tan(0.0) == 0.0);
try expect(tan(-0.0) == -0.0);
try expect(math.isNan(tan(math.inf(f64))));
try expect(math.isNan(tan(-math.inf(f64))));
try expect(math.isNan(tan(math.nan(f64))));
}