Merge pull request #25691 from GasInfinity-Forks/x86_16-gcc

feat: init x86_16 arch via CBE

lib/compiler_rt/int_from_float.zig

@@ -74,7 +74,7 @@ pub inline fn bigIntFromFloat(comptime signedness: std.builtin.Signedness, resul
     const parts = math.frexp(a);
     const significand_bits_adjusted_to_handle_smin = @as(i32, significand_bits) +
         @intFromBool(signedness == .signed and parts.exponent == 32 * result.len);
-    const exponent = @max(parts.exponent - significand_bits_adjusted_to_handle_smin, 0);
+    const exponent: usize = @intCast(@max(parts.exponent - significand_bits_adjusted_to_handle_smin, 0));
     const int: I = @intFromFloat(switch (exponent) {
         0 => a,
         else => math.ldexp(parts.significand, significand_bits_adjusted_to_handle_smin),

lib/std/Io/Writer.zig

@@ -604,7 +604,7 @@ pub fn print(w: *Writer, comptime fmt: []const u8, args: anytype) Error!void {
         @compileError("32 arguments max are supported per format call");
     }
 
-    @setEvalBranchQuota(fmt.len * 1000);
+    @setEvalBranchQuota(@as(comptime_int, fmt.len) * 1000); // NOTE: We're upcasting as 16-bit usize overflows.
     comptime var arg_state: std.fmt.ArgState = .{ .args_len = fields_info.len };
     comptime var i = 0;
     comptime var literal: []const u8 = "";

lib/std/Target.zig

@@ -1083,7 +1083,7 @@ pub fn toElfMachine(target: *const Target) std.elf.EM {
         .sparc => if (target.cpu.has(.sparc, .v9)) .SPARC32PLUS else .SPARC,
         .sparc64 => .SPARCV9,
         .ve => .VE,
-        .x86 => .@"386",
+        .x86_16, .x86 => .@"386",
         .x86_64 => .X86_64,
         .xcore => .XCORE,
         .xtensa, .xtensaeb => .XTENSA,
@@ -1154,6 +1154,7 @@ pub fn toCoffMachine(target: *const Target) std.coff.IMAGE.FILE.MACHINE {
         .ve,
         .wasm32,
         .wasm64,
+        .x86_16,
         .xcore,
         .xtensa,
         .xtensaeb,
@@ -1376,6 +1377,7 @@ pub const Cpu = struct {
         ve,
         wasm32,
         wasm64,
+        x86_16,
         x86,
         x86_64,
         xcore,
@@ -1467,7 +1469,7 @@ pub const Cpu = struct {
                 .spirv32, .spirv64 => .spirv,
                 .ve => .ve,
                 .wasm32, .wasm64 => .wasm,
-                .x86, .x86_64 => .x86,
+                .x86_16, .x86, .x86_64 => .x86,
                 .xcore => .xcore,
                 .xtensa, .xtensaeb => .xtensa,
             };
@@ -1475,7 +1477,7 @@ pub const Cpu = struct {
 
         pub inline fn isX86(arch: Arch) bool {
             return switch (arch) {
-                .x86, .x86_64 => true,
+                .x86_16, .x86, .x86_64 => true,
                 else => false,
             };
         }
@@ -1669,6 +1671,7 @@ pub const Cpu = struct {
                 .ve,
                 .wasm32,
                 .wasm64,
+                .x86_16,
                 .x86,
                 .x86_64,
                 .xcore,
@@ -1789,6 +1792,12 @@ pub const Cpu = struct {
                 .x86_interrupt,
                 => &.{.x86},
 
+                .x86_16_cdecl,
+                .x86_16_stdcall,
+                .x86_16_regparmcall,
+                .x86_16_interrupt,
+                => &.{.x86_16},
+
                 .aarch64_aapcs,
                 .aarch64_aapcs_darwin,
                 .aarch64_aapcs_win,
@@ -1971,6 +1980,7 @@ pub const Cpu = struct {
                 .riscv64, .riscv64be => &riscv.cpu.generic_rv64,
                 .sparc64 => &sparc.cpu.v9, // SPARC can only be 64-bit from v9 and up.
                 .wasm32, .wasm64 => &wasm.cpu.mvp,
+                .x86_16 => &x86.cpu.i86,
                 .x86 => &x86.cpu.i386,
                 .x86_64 => &x86.cpu.x86_64,
                 inline else => |a| &@field(Target, @tagName(a.family())).cpu.generic,
@@ -2237,7 +2247,10 @@ pub fn supportsAddressSpace(
 
     return switch (address_space) {
         .generic => true,
-        .fs, .gs, .ss => (arch == .x86_64 or arch == .x86) and (context == null or context == .pointer),
+        .fs, .gs, .ss => (arch == .x86_64 or arch == .x86 or arch == .x86_16) and (context == null or context == .pointer),
+        // Technically x86 can use segmentation...
+        .far => (arch == .x86_16),
+
         .flash, .flash1, .flash2, .flash3, .flash4, .flash5 => arch == .avr, // TODO this should also check how many flash banks the cpu has
         .cog, .hub => arch == .propeller,
         .lut => arch == .propeller and std.Target.propeller.featureSetHas(target.cpu.features, .p2),
@@ -2800,6 +2813,7 @@ pub fn ptrBitWidth_arch_abi(cpu_arch: Cpu.Arch, abi: Abi) u16 {
     return switch (cpu_arch) {
         .avr,
         .msp430,
+        .x86_16,
         => 16,
 
         .arc,
@@ -3013,7 +3027,7 @@ pub fn cTypeByteSize(t: *const Target, c_type: CType) u16 {
 pub fn cTypeBitSize(target: *const Target, c_type: CType) u16 {
     switch (target.os.tag) {
         .freestanding, .other => switch (target.cpu.arch) {
-            .msp430 => switch (c_type) {
+            .msp430, .x86_16 => switch (c_type) {
                 .char => return 8,
                 .short, .ushort, .int, .uint => return 16,
                 .float, .long, .ulong => return 32,
@@ -3369,6 +3383,7 @@ pub fn cTypeAlignment(target: *const Target, c_type: CType) u16 {
         std.math.ceilPowerOfTwoAssert(u16, (cTypeBitSize(target, c_type) + 7) / 8),
         @as(u16, switch (target.cpu.arch) {
             .msp430,
+            .x86_16,
             => 2,
 
             .arc,
@@ -3476,7 +3491,7 @@ pub fn cTypePreferredAlignment(target: *const Target, c_type: CType) u16 {
     return @min(
         std.math.ceilPowerOfTwoAssert(u16, (cTypeBitSize(target, c_type) + 7) / 8),
         @as(u16, switch (target.cpu.arch) {
-            .msp430 => 2,
+            .x86_16, .msp430 => 2,
 
             .arc,
             .arceb,
@@ -3548,7 +3563,7 @@ pub fn cMaxIntAlignment(target: *const Target) u16 {
     return switch (target.cpu.arch) {
         .avr => 1,
 
-        .msp430 => 2,
+        .msp430, .x86_16 => 2,
 
         .arc,
         .arceb,
@@ -3625,6 +3640,7 @@ pub fn cCallingConvention(target: *const Target) ?std.builtin.CallingConvention
             .windows, .uefi => .{ .x86_win = .{} },
             else => .{ .x86_sysv = .{} },
         },
+        .x86_16 => .{ .x86_16_cdecl = .{} },
         .aarch64, .aarch64_be => if (target.os.tag.isDarwin())
             .{ .aarch64_aapcs_darwin = .{} }
         else switch (target.os.tag) {

lib/std/Target/x86.zig

@@ -3081,6 +3081,11 @@ pub const cpu = struct {
             .xsaveopt,
         }),
     };
+    pub const @"i86": CpuModel = .{
+        .name = "i86",
+        .llvm_name = null,
+        .features = featureSet(&[_]Feature{}),
+    };
     pub const @"i386": CpuModel = .{
         .name = "i386",
         .llvm_name = "i386",

lib/std/builtin.zig

@@ -223,6 +223,13 @@ pub const CallingConvention = union(enum(u8)) {
     x86_vectorcall: CommonOptions,
     x86_interrupt: CommonOptions,
 
+    // Calling conventions for the `x86_16` architecture.
+
+    x86_16_cdecl: CommonOptions,
+    x86_16_stdcall: CommonOptions,
+    x86_16_regparmcall: CommonOptions,
+    x86_16_interrupt: CommonOptions,
+
     // Calling conventions for the `aarch64` and `aarch64_be` architectures.
     aarch64_aapcs: CommonOptions,
     aarch64_aapcs_darwin: CommonOptions,
@@ -523,6 +530,10 @@ pub const AddressSpace = enum(u5) {
     fs,
     ss,
 
+    // x86_16 extra address spaces.
+    /// Allows addressing the entire address space by storing both segment and offset.
+    far,
+
     // GPU address spaces.
     global,
     constant,

lib/std/builtin/assembly.zig

@@ -1,5 +1,5 @@
 pub const Clobbers = switch (@import("builtin").cpu.arch) {
-    .x86, .x86_64 => packed struct {
+    .x86_16, .x86, .x86_64 => packed struct {
         /// Whether the inline assembly code may perform stores to memory
         /// addresses other than those derived from input pointer provenance.
         memory: bool = false,

lib/std/os/windows.zig

@@ -4634,25 +4634,28 @@ pub const TEB = extern struct {
 };
 
 comptime {
-    // Offsets taken from WinDbg info and Geoff Chappell[1] (RIP)
-    // [1]: https://www.geoffchappell.com/studies/windows/km/ntoskrnl/inc/api/pebteb/teb/index.htm
-    assert(@offsetOf(TEB, "NtTib") == 0x00);
-    if (@sizeOf(usize) == 4) {
-        assert(@offsetOf(TEB, "EnvironmentPointer") == 0x1C);
-        assert(@offsetOf(TEB, "ClientId") == 0x20);
-        assert(@offsetOf(TEB, "ActiveRpcHandle") == 0x28);
-        assert(@offsetOf(TEB, "ThreadLocalStoragePointer") == 0x2C);
-        assert(@offsetOf(TEB, "ProcessEnvironmentBlock") == 0x30);
-        assert(@offsetOf(TEB, "LastErrorValue") == 0x34);
-        assert(@offsetOf(TEB, "TlsSlots") == 0xe10);
-    } else if (@sizeOf(usize) == 8) {
-        assert(@offsetOf(TEB, "EnvironmentPointer") == 0x38);
-        assert(@offsetOf(TEB, "ClientId") == 0x40);
-        assert(@offsetOf(TEB, "ActiveRpcHandle") == 0x50);
-        assert(@offsetOf(TEB, "ThreadLocalStoragePointer") == 0x58);
-        assert(@offsetOf(TEB, "ProcessEnvironmentBlock") == 0x60);
-        assert(@offsetOf(TEB, "LastErrorValue") == 0x68);
-        assert(@offsetOf(TEB, "TlsSlots") == 0x1480);
+    // XXX: Without this check we cannot use `std.Io.Writer` on 16-bit platforms. `std.fmt.bufPrint` will hit the unreachable in `PEB.GdiHandleBuffer` without this guard.
+    if (builtin.os.tag == .windows) {
+        // Offsets taken from WinDbg info and Geoff Chappell[1] (RIP)
+        // [1]: https://www.geoffchappell.com/studies/windows/km/ntoskrnl/inc/api/pebteb/teb/index.htm
+        assert(@offsetOf(TEB, "NtTib") == 0x00);
+        if (@sizeOf(usize) == 4) {
+            assert(@offsetOf(TEB, "EnvironmentPointer") == 0x1C);
+            assert(@offsetOf(TEB, "ClientId") == 0x20);
+            assert(@offsetOf(TEB, "ActiveRpcHandle") == 0x28);
+            assert(@offsetOf(TEB, "ThreadLocalStoragePointer") == 0x2C);
+            assert(@offsetOf(TEB, "ProcessEnvironmentBlock") == 0x30);
+            assert(@offsetOf(TEB, "LastErrorValue") == 0x34);
+            assert(@offsetOf(TEB, "TlsSlots") == 0xe10);
+        } else if (@sizeOf(usize) == 8) {
+            assert(@offsetOf(TEB, "EnvironmentPointer") == 0x38);
+            assert(@offsetOf(TEB, "ClientId") == 0x40);
+            assert(@offsetOf(TEB, "ActiveRpcHandle") == 0x50);
+            assert(@offsetOf(TEB, "ThreadLocalStoragePointer") == 0x58);
+            assert(@offsetOf(TEB, "ProcessEnvironmentBlock") == 0x60);
+            assert(@offsetOf(TEB, "LastErrorValue") == 0x68);
+            assert(@offsetOf(TEB, "TlsSlots") == 0x1480);
+        }
     }
 }
 

lib/std/zig/system.zig

@@ -374,6 +374,11 @@ pub fn resolveTargetQuery(query: Target.Query) DetectError!Target {
     // However, the "mode" flags can be used as overrides, so if the user explicitly
     // sets one of them, that takes precedence.
     switch (query_cpu_arch) {
+        .x86_16 => {
+            cpu.features.addFeature(
+                @intFromEnum(Target.x86.Feature.@"16bit_mode"),
+            );
+        },
         .x86 => {
             if (!Target.x86.featureSetHasAny(query.cpu_features_add, .{
                 .@"16bit_mode", .@"32bit_mode",

lib/std/zig/target.zig

@@ -472,8 +472,9 @@ fn eqlIgnoreCase(ignore_case: bool, a: []const u8, b: []const u8) bool {
     }
 }
 
-pub fn intByteSize(target: *const std.Target, bits: u16) u19 {
-    return std.mem.alignForward(u19, @intCast((@as(u17, bits) + 7) / 8), intAlignment(target, bits));
+pub fn intByteSize(target: *const std.Target, bits: u16) u16 {
+    const previous_aligned = std.mem.alignBackward(u16, bits, 8);
+    return std.mem.alignForward(u16, @divExact(previous_aligned, 8) + @intFromBool(previous_aligned != bits), intAlignment(target, bits));
 }
 
 pub fn intAlignment(target: *const std.Target, bits: u16) u16 {

lib/zig.h

@@ -74,6 +74,9 @@
 #elif defined (__x86_64__) || (defined(zig_msvc) && defined(_M_X64))
 #define zig_x86_64
 #define zig_x86
+#elif defined(__I86__)
+#define zig_x86_16
+#define zig_x86
 #endif
 
 #if defined(zig_msvc) || __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
@@ -400,6 +403,8 @@
 #define zig_trap() __asm__ volatile("j 0x2")
 #elif defined(zig_sparc)
 #define zig_trap() __asm__ volatile("illtrap")
+#elif defined(zig_x86_16)
+#define zig_trap() __asm__ volatile("int $0x3")
 #elif defined(zig_x86)
 #define zig_trap() __asm__ volatile("ud2")
 #else
@@ -4219,7 +4224,7 @@ static inline void zig_loongarch_cpucfg(uint32_t word, uint32_t* result) {
 #endif
 }
 
-#elif defined(zig_x86)
+#elif defined(zig_x86) && !defined(zig_x86_16)
 
 static inline void zig_x86_cpuid(uint32_t leaf_id, uint32_t subid, uint32_t* eax, uint32_t* ebx, uint32_t* ecx, uint32_t* edx) {
 #if defined(zig_msvc)

src/Sema.zig

@@ -9034,6 +9034,7 @@ pub fn handleExternLibName(
 /// Any calling conventions not included here are either not yet verified to work with variadic
 /// functions or there are no more other calling conventions that support variadic functions.
 const calling_conventions_supporting_var_args = [_]std.builtin.CallingConvention.Tag{
+    .x86_16_cdecl,
     .x86_64_sysv,
     .x86_64_x32,
     .x86_64_win,

src/Zcu.zig

@@ -4406,6 +4406,10 @@ pub fn callconvSupported(zcu: *Zcu, cc: std.builtin.CallingConvention) union(enu
                 }
             }
             break :ok switch (cc) {
+                .x86_16_cdecl,
+                .x86_16_stdcall,
+                .x86_16_regparmcall,
+                .x86_16_interrupt,
                 .x86_64_sysv,
                 .x86_64_win,
                 .x86_64_vectorcall,

src/codegen/c.zig

@@ -8055,9 +8055,11 @@ fn toCallingConvention(cc: std.builtin.CallingConvention, zcu: *Zcu) ?[]const u8
     return switch (cc) {
         .auto, .naked => null,
 
+        .x86_16_cdecl => "cdecl",
+        .x86_16_regparmcall => "regparmcall",
         .x86_64_sysv, .x86_sysv => "sysv_abi",
         .x86_64_win, .x86_win => "ms_abi",
-        .x86_stdcall => "stdcall",
+        .x86_16_stdcall, .x86_stdcall => "stdcall",
         .x86_fastcall => "fastcall",
         .x86_thiscall => "thiscall",
 
@@ -8127,6 +8129,7 @@ fn toCallingConvention(cc: std.builtin.CallingConvention, zcu: *Zcu) ?[]const u8
         .csky_interrupt,
         .m68k_interrupt,
         .msp430_interrupt,
+        .x86_16_interrupt,
         .x86_interrupt,
         .x86_64_interrupt,
         => "interrupt",

src/codegen/llvm.zig

@@ -117,6 +117,7 @@ pub fn targetTriple(allocator: Allocator, target: *const std.Target) ![]const u8
         .propeller,
         .sh,
         .sheb,
+        .x86_16,
         .xtensaeb,
         => unreachable, // Gated by hasLlvmSupport().
     };
@@ -493,6 +494,7 @@ pub fn dataLayout(target: *const std.Target) []const u8 {
         .propeller,
         .sh,
         .sheb,
+        .x86_16,
         .xtensaeb,
         => unreachable, // Gated by hasLlvmSupport().
     };
@@ -11902,6 +11904,10 @@ fn toLlvmCallConvTag(cc_tag: std.builtin.CallingConvention.Tag, target: *const s
 
         // All the calling conventions which LLVM does not have a general representation for.
         // Note that these are often still supported through the `cCallingConvention` path above via `ccc`.
+        .x86_16_cdecl,
+        .x86_16_stdcall,
+        .x86_16_regparmcall,
+        .x86_16_interrupt,
         .x86_sysv,
         .x86_win,
         .x86_thiscall_mingw,
@@ -13131,6 +13137,7 @@ pub fn initializeLLVMTarget(arch: std.Target.Cpu.Arch) void {
         .propeller,
         .sh,
         .sheb,
+        .x86_16,
         .xtensaeb,
         => unreachable,
     }

src/codegen/spirv/Module.zig

@@ -927,6 +927,7 @@ pub fn storageClass(module: *Module, as: std.builtin.AddressSpace) spec.StorageC
         .gs,
         .fs,
         .ss,
+        .far,
         .param,
         .flash,
         .flash1,

src/target.zig

@@ -227,6 +227,7 @@ pub fn hasLlvmSupport(target: *const std.Target, ofmt: std.Target.ObjectFormat)
         .propeller,
         .sh,
         .sheb,
+        .x86_16,
         .xtensaeb,
         => false,
     };