llvm: fix @max/@min of unsupported float types

Closes #15611
This commit is contained in:
Jacob Young 2023-05-09 22:23:44 -04:00
parent 2d2d79a05b
commit c5b96c7447
6 changed files with 55 additions and 23 deletions

View File

@ -37,6 +37,7 @@ pub const sqrt2 = 1.414213562373095048801688724209698079;
/// 1/sqrt(2)
pub const sqrt1_2 = 0.707106781186547524400844362104849039;
pub const floatBits = @import("math/float.zig").floatBits;
pub const floatExponentBits = @import("math/float.zig").floatExponentBits;
pub const floatMantissaBits = @import("math/float.zig").floatMantissaBits;
pub const floatFractionalBits = @import("math/float.zig").floatFractionalBits;

View File

@ -4,16 +4,17 @@ const expect = std.testing.expect;
/// Returns a value with the magnitude of `magnitude` and the sign of `sign`.
pub fn copysign(magnitude: anytype, sign: @TypeOf(magnitude)) @TypeOf(magnitude) {
const T = @TypeOf(magnitude);
const TBits = std.meta.Int(.unsigned, @typeInfo(T).Float.bits);
const sign_bit_mask = @as(TBits, 1) << (@bitSizeOf(T) - 1);
const mag = @bitCast(TBits, magnitude) & ~sign_bit_mask;
const sgn = @bitCast(TBits, sign) & sign_bit_mask;
return @bitCast(T, mag | sgn);
const bits = math.floatBits(@TypeOf(magnitude));
const FBits = @Type(.{ .Float = .{ .bits = bits } });
const TBits = @Type(.{ .Int = .{ .signedness = .unsigned, .bits = bits } });
const sign_bit_mask = @as(TBits, 1) << (bits - 1);
const mag = @bitCast(TBits, @as(FBits, magnitude)) & ~sign_bit_mask;
const sgn = @bitCast(TBits, @as(FBits, sign)) & sign_bit_mask;
return @bitCast(FBits, mag | sgn);
}
test "math.copysign" {
inline for ([_]type{ f16, f32, f64, f80, f128 }) |T| {
inline for ([_]type{ f16, f32, f64, f80, f128, c_longdouble, comptime_float }) |T| {
try expect(copysign(@as(T, 1.0), @as(T, 1.0)) == 1.0);
try expect(copysign(@as(T, 2.0), @as(T, -2.0)) == -2.0);
try expect(copysign(@as(T, -3.0), @as(T, 3.0)) == 3.0);

View File

@ -4,21 +4,29 @@ const expect = std.testing.expect;
/// Creates a raw "1.0" mantissa for floating point type T. Used to dedupe f80 logic.
inline fn mantissaOne(comptime T: type) comptime_int {
return if (@typeInfo(T).Float.bits == 80) 1 << floatFractionalBits(T) else 0;
return 1 << floatFractionalBits(T) & ((1 << floatMantissaBits(T)) - 1);
}
/// Creates floating point type T from an unbiased exponent and raw mantissa.
inline fn reconstructFloat(comptime T: type, comptime exponent: comptime_int, comptime mantissa: comptime_int) T {
const TBits = @Type(.{ .Int = .{ .signedness = .unsigned, .bits = @bitSizeOf(T) } });
const FBits = @Type(.{ .Float = .{ .bits = floatBits(T) } });
const TBits = @Type(.{ .Int = .{ .signedness = .unsigned, .bits = floatBits(T) } });
const biased_exponent = @as(TBits, exponent + floatExponentMax(T));
return @bitCast(T, (biased_exponent << floatMantissaBits(T)) | @as(TBits, mantissa));
return @bitCast(FBits, (biased_exponent << floatMantissaBits(T)) | @as(TBits, mantissa));
}
/// Returns the number of bits in floating point type T.
pub inline fn floatBits(comptime T: type) comptime_int {
return switch (@typeInfo(T)) {
.Float => |info| info.bits,
.ComptimeFloat => 128,
else => @compileError(@typeName(T) ++ " is not a floating point type"),
};
}
/// Returns the number of bits in the exponent of floating point type T.
pub inline fn floatExponentBits(comptime T: type) comptime_int {
comptime assert(@typeInfo(T) == .Float);
return switch (@typeInfo(T).Float.bits) {
return switch (floatBits(T)) {
16 => 5,
32 => 8,
64 => 11,
@ -30,9 +38,7 @@ pub inline fn floatExponentBits(comptime T: type) comptime_int {
/// Returns the number of bits in the mantissa of floating point type T.
pub inline fn floatMantissaBits(comptime T: type) comptime_int {
comptime assert(@typeInfo(T) == .Float);
return switch (@typeInfo(T).Float.bits) {
return switch (floatBits(T)) {
16 => 10,
32 => 23,
64 => 52,
@ -44,12 +50,10 @@ pub inline fn floatMantissaBits(comptime T: type) comptime_int {
/// Returns the number of fractional bits in the mantissa of floating point type T.
pub inline fn floatFractionalBits(comptime T: type) comptime_int {
comptime assert(@typeInfo(T) == .Float);
// standard IEEE floats have an implicit 0.m or 1.m integer part
// f80 is special and has an explicitly stored bit in the MSB
// this function corresponds to `MANT_DIG - 1' from C
return switch (@typeInfo(T).Float.bits) {
return switch (floatBits(T)) {
16 => 10,
32 => 23,
64 => 52,
@ -101,6 +105,7 @@ test "float bits" {
inline for ([_]type{ f16, f32, f64, f80, f128, c_longdouble }) |T| {
// (1 +) for the sign bit, since it is separate from the other bits
const size = 1 + floatExponentBits(T) + floatMantissaBits(T);
try expect(floatBits(T) == size);
try expect(@bitSizeOf(T) == size);
// for machine epsilon, assert expmin <= -prec <= expmax

View File

@ -2,13 +2,13 @@ const math = @import("../math.zig");
/// Returns the nan representation for type T.
pub inline fn nan(comptime T: type) T {
return switch (@typeInfo(T).Float.bits) {
return switch (math.floatBits(T)) {
16 => math.nan_f16,
32 => math.nan_f32,
64 => math.nan_f64,
80 => math.nan_f80,
128 => math.nan_f128,
else => @compileError("unreachable"),
else => @compileError("unknown floating point type " ++ @typeName(T)),
};
}

View File

@ -7034,7 +7034,7 @@ pub const FuncGen = struct {
const rhs = try self.resolveInst(bin_op.rhs);
const scalar_ty = self.air.typeOfIndex(inst).scalarType();
if (scalar_ty.isAnyFloat()) return self.builder.buildMinNum(lhs, rhs, "");
if (scalar_ty.isAnyFloat()) return self.buildFloatOp(.fmin, scalar_ty, 2, .{ lhs, rhs });
if (scalar_ty.isSignedInt()) return self.builder.buildSMin(lhs, rhs, "");
return self.builder.buildUMin(lhs, rhs, "");
}
@ -7045,7 +7045,7 @@ pub const FuncGen = struct {
const rhs = try self.resolveInst(bin_op.rhs);
const scalar_ty = self.air.typeOfIndex(inst).scalarType();
if (scalar_ty.isAnyFloat()) return self.builder.buildMaxNum(lhs, rhs, "");
if (scalar_ty.isAnyFloat()) return self.buildFloatOp(.fmax, scalar_ty, 2, .{ lhs, rhs });
if (scalar_ty.isSignedInt()) return self.builder.buildSMax(lhs, rhs, "");
return self.builder.buildUMax(lhs, rhs, "");
}

View File

@ -96,6 +96,31 @@ test "@min for vectors" {
comptime try S.doTheTest();
}
test "@min/max for floats" {
if (builtin.zig_backend == .stage2_wasm) return error.SkipZigTest; // TODO
if (builtin.zig_backend == .stage2_x86_64) return error.SkipZigTest; // TODO
if (builtin.zig_backend == .stage2_aarch64) return error.SkipZigTest; // TODO
if (builtin.zig_backend == .stage2_arm) return error.SkipZigTest; // TODO
if (builtin.zig_backend == .stage2_sparc64) return error.SkipZigTest; // TODO
const S = struct {
fn doTheTest(comptime T: type) !void {
var x: T = -3.14;
var y: T = 5.27;
try expectEqual(x, @min(x, y));
try expectEqual(x, @min(y, x));
try expectEqual(y, @max(x, y));
try expectEqual(y, @max(y, x));
}
};
inline for (.{ f16, f32, f64, f80, f128, c_longdouble }) |T| {
try S.doTheTest(T);
comptime try S.doTheTest(T);
}
comptime try S.doTheTest(comptime_float);
}
test "@min/@max on lazy values" {
const A = extern struct { u8_4: [4]u8 };
const B = extern struct { u8_16: [16]u8 };