native: initial support for apple-m1 (#14795)

.github/workflows/ci.yml

@@ -151,7 +151,7 @@ jobs:
           [ "$(stat -c %s leaks.txt)" = "0" ]
 
   macos:
-    runs-on: macOS-latest
+    runs-on: macOS-12
     if: github.event_name != 'push' || github.event.ref == 'refs/heads/master' || github.event.repository.full_name != 'vlang/v'
     timeout-minutes: 121
     env:

vlib/v/builder/nativebuilder/nativebuilder.v

@@ -39,6 +39,15 @@ pub fn build_native(mut b builder.Builder, v_files []string, out_file string) {
 	}
 	b.front_and_middle_stages(nvf) or { return }
 	util.timing_start('Native GEN')
+	if b.pref.arch == ._auto {
+		$if amd64 {
+			b.pref.arch = .amd64
+		} $else $if arm64 {
+			b.pref.arch = .arm64
+		} $else {
+			eprintln('Error: Only arm64 and amd64 are supported by V')
+		}
+	}
 	b.stats_lines, b.stats_bytes = native.gen(b.parsed_files, b.table, out_file, b.pref)
 	util.timing_measure('Native GEN')
 }

vlib/v/gen/native/amd64.v

@@ -76,6 +76,9 @@ fn (mut g Gen) neg(reg Register) {
 }
 
 fn (mut g Gen) cmp(reg Register, size Size, val i64) {
+	if g.pref.arch != .amd64 {
+		panic('cmp')
+	}
 	// Second byte depends on the size of the value
 	match size {
 		._8 {
@@ -349,6 +352,9 @@ fn (mut g Gen) movabs(reg Register, val i64) {
 }
 
 fn (mut g Gen) mov_reg_to_var(var_offset int, reg Register) {
+	if g.pref.arch != .amd64 {
+		panic('invalid arch for mov_reg_to_var')
+	}
 	// 89 7d fc     mov DWORD PTR [rbp-0x4],edi
 	match reg {
 		.rax {
@@ -458,8 +464,14 @@ fn (mut g Gen) cdq() {
 }
 
 pub fn (mut g Gen) ret() {
+	if g.pref.arch == .amd64 {
 		g.write8(0xc3)
 		g.println('ret')
+	} else if g.pref.arch == .arm64 {
+		g.write32(0xd65f03c0)
+	} else {
+		panic('invalid arch')
+	}
 }
 
 pub fn (mut g Gen) push(reg Register) {
@@ -1958,9 +1970,11 @@ fn (mut g Gen) fn_decl_amd64(node ast.FnDecl) {
 	g.ret()
 }
 
+/*
 pub fn (mut x Amd64) allocate_var(name string, size int, initial_val int) {
 	// do nothing as interface call is crashing
 }
+*/
 
 pub fn (mut g Gen) allocate_array(name string, size int, items int) int {
 	pos := g.allocate_var(name, size, items)
@@ -1969,6 +1983,10 @@ pub fn (mut g Gen) allocate_array(name string, size int, items int) int {
 }
 
 pub fn (mut g Gen) allocate_var(name string, size int, initial_val int) int {
+	if g.pref.arch == .arm64 {
+		// TODO
+		return 0
+	}
 	// `a := 3`  => `mov DWORD [rbp-0x4],0x3`
 	match size {
 		1 {

vlib/v/gen/native/arm64.v

@@ -42,7 +42,7 @@ mut:
 	// arm64 specific stuff for code generation
 }
 
-pub fn (mut x Arm64) allocate_var(name string, size int, initial_val int) {
+fn (mut x Arm64) allocate_var(name string, size int, initial_val int) {
 	eprintln('TODO: allocating var on arm64 ($name) = $size = $initial_val')
 }
 
@@ -54,11 +54,8 @@ fn (mut g Gen) mov_arm(reg Arm64Register, val u64) {
 	// println(x & ~(m << 16))
 	// g.write32(0x777777)
 	r := int(reg)
-	if r == 0 && val == 1 {
+	if r >= 0 && r <= 16 {
-		g.write32(0xd2800020)
+		g.write32(int(u32(0xd2800000 + u32(r) + (u32(val) << 5))))
-		g.println('mov x0, 1')
-	} else if r >= 0 && r <= 16 {
-		g.write32(int(u32(0xd2800000 + int(r) + int(val)) << 5))
 		g.println('mov x$r, $val')
 	} else {
 		g.n_error('mov_arm unsupported values')
@@ -82,17 +79,106 @@ fn (mut g Gen) neg_arm(r Arm64Register) {
 }
 
 fn (mut g Gen) neg_regs_arm(a Arm64Register, b Arm64Register) {
-	if int(a) < 0x0f && int(b) < 0x0f {
+	if u32(a) < 0x0f && u32(b) < 0x0f {
-		g.write32(0xe2600000 | int(a) << 16 | int(b) << 12)
+		g.write32(int(0xe2600000 | (u32(a) << 16) | u32(b) << 12))
 		g.println('neg $a, $b')
 	} else {
 		g.n_error('unhandled neg $a, $b')
 	}
 }
 
+fn (mut g Gen) sub_sp(v int) {
+	if g.pref.arch != .arm64 {
+		g.n_error('sub_sp is arm64-specifig')
+		return
+	}
+	// this is for 0x20 only
+	if v < 0 {
+		g.write32(i32(0x910083ff)) // add sp, X
+	} else {
+		g.write32(i32(0xd10083ff)) // sub sp, X
+	}
+}
+
 pub fn (mut g Gen) fn_decl_arm64(node ast.FnDecl) {
 	g.gen_arm64_helloworld()
-	// TODO
+	/*
+	0x100003f6c      ff8300d1       sub sp, sp, 0x20           ; [00] -r-x section size 52 named 0.__TEXT.__text
+            0x100003f70      fd7b01a9       stp x29, x30, [sp, 0x10]
+            0x100003f74      fd430091       add x29, sp, 0x10
+            0x100003f78      bfc31fb8       stur wzr, [x29, -4]
+            0x100003f7c      68008052       mov w8, 3
+            0x100003f80      e80b00b9       str w8, [sp, 8]
+            0x100003f84      00000090       adrp x0, 0x100003000
+            0x100003f88      00b03e91       add x0, x0, 0xfac
+            0x100003f8c      05000094       bl sym.imp.puts            ;[1]
+            0x100003f90      e00b40b9       ldr w0, [sp, 8]            ; 5
+            0x100003f94      fd7b41a9       ldp x29, x30, [sp, 0x10]
+            0x100003f98      ff830091       add sp, sp, 0x20
+            0x100003f9c      c0035fd6       ret
+	*/
+
+	/*
+	/*
+	g.push(.rbp)
+	g.mov_rbp_rsp()
+*/
+	locals_count := node.scope.objects.len + node.params.len + node.defer_stmts.len
+	g.stackframe_size = (locals_count * 8) + 0x10
+//	g.sub8(.rsp, g.stackframe_size)
+	g.sub_sp(32)
+
+	// Copy values from registers to local vars (calling convention)
+	mut offset := 0
+	for i in 0 .. node.params.len {
+		name := node.params[i].name
+		// TODO optimize. Right now 2 mov's are used instead of 1.
+		g.allocate_var(name, 4, 0)
+		// `mov DWORD PTR [rbp-0x4],edi`
+		offset += 4
+		g.mov_reg_to_var(offset, native.fn_arg_registers[i])
+	}
+	// define defer vars
+	for i in 0 .. node.defer_stmts.len {
+		name := '_defer$i'
+		g.allocate_var(name, 8, 0)
+	}
+	//
+	g.stmts(node.stmts)
+	is_main := node.name == 'main.main'
+	if is_main {
+		// println('end of main: gen exit')
+		zero := ast.IntegerLiteral{}
+		g.gen_exit(zero)
+		g.ret()
+		return
+	}
+	// g.leave()
+	g.labels.addrs[0] = g.pos()
+	g.println('; label 0: return')
+/*
+	if g.defer_stmts.len != 0 {
+		// save return value
+		g.push(.rax)
+		for defer_stmt in g.defer_stmts.reverse() {
+			defer_var := g.get_var_offset('_defer$defer_stmt.idx_in_fn')
+			g.mov_var_to_reg(.rax, defer_var)
+			g.cmp_zero(.rax)
+			label := g.labels.new_label()
+			jump_addr := g.cjmp(.je)
+			g.labels.patches << LabelPatch{
+				id: label
+				pos: jump_addr
+			}
+			g.stmts(defer_stmt.stmts)
+			g.labels.addrs[label] = g.pos()
+		}
+		//g.pop(.rax)
+	}
+*/
+	g.sub_sp(-32)
+	g.ret()
+	*/
 }
 
 pub fn (mut g Gen) call_fn_arm64(node ast.CallExpr) {
@@ -143,6 +229,11 @@ fn (mut g Gen) gen_arm64_helloworld() {
 		g.mov_arm(.x8, 64) // write (linux-arm64)
 		g.svc()
 	} else {
+		g.mov_arm(.x0, 1)
+		g.adr(.x1, 0x10 + 4)
+		g.mov_arm(.x2, 13)
+		g.mov_arm(.x16, 4) // write
+		g.svc()
 		g.mov_arm(.x0, 0)
 		g.mov_arm(.x16, 1)
 		g.svc()
@@ -167,8 +258,13 @@ fn (mut g Gen) bl() {
 }
 
 fn (mut g Gen) svc() {
+	if g.pref.os == .linux {
 		g.write32(0xd4001001)
 		g.println('svc 0x80')
+	} else {
+		g.write32(0xd4000001)
+		g.println('svc 0')
+	}
 }
 
 pub fn (mut c Arm64) gen_exit(mut g Gen, expr ast.Expr) {

vlib/v/gen/native/gen.v

@@ -49,6 +49,9 @@ mut:
 	strs                 []String
 	labels               &LabelTable
 	defer_stmts          []ast.DeferStmt
+	// macho specific
+	macho_ncmds   int
+	macho_cmdsize int
 }
 
 enum RelocType {
@@ -114,6 +117,20 @@ fn get_backend(arch pref.Arch) ?CodeGen {
 				g: 0
 			}
 		}
+		._auto {
+			$if amd64 {
+				return Amd64{
+					g: 0
+				}
+			} $else $if arm64 {
+				return Arm64{
+					g: 0
+				}
+			} $else {
+				eprintln('-native only have amd64 and arm64 codegens')
+				exit(1)
+			}
+		}
 		else {}
 	}
 	return error('unsupported architecture')

vlib/v/gen/native/macho.v

@@ -4,21 +4,24 @@
 module native
 
 const (
-	s_attr_some_instructions = 0x00000400
+	s_attr_some_instructions = 0x0400
 	s_attr_pure_instructions = 0x80000000
-	s_attr_ext_reloc         = 0x00000200
+	s_attr_ext_reloc         = 0x0200
-	s_attr_loc_reloc         = 0x00000100
+	s_attr_loc_reloc         = 0x0100
-	//
 	macho_symcmd_size        = 0x18
 	macho_d_size             = 0x50
 	mh_object                = 1
 	mh_execute               = 2
+	lc_dyld_chained_fixups   = 0x80000034
+	lc_dyld_exports_trie     = 0x80000033
+	lc_dyld_info_only        = 0x80000022
 	lc_dysymtab              = 0xb
 	lc_load_dylib            = 0xc
 	lc_load_dylinker         = 0xe
 	lc_main                  = 0x80000028
 	lc_segment_64            = 0x19
 	lc_symtab                = 0x2
+	base_addr                = i64(0x1_0000_0000)
 )
 
 struct Symbol {
@@ -41,31 +44,53 @@ struct Reloc {
 }
 
 fn (mut g Gen) macho_segment64_pagezero() {
-	g.write32(native.lc_segment_64) // LC_SEGMENT_64
+	g.macho_add_loadcommand(native.lc_segment_64, 72)
-	g.write32(72) // cmdsize
 	g.write_string_with_padding('__PAGEZERO', 16) // section name
 	g.write64(0) // vmaddr
-	g.write64(0x1000) // vmsize
+	// XXX g.write64(g.get_pagesize()) // vmsize
+	g.write64(native.base_addr) // vmsize
 	g.write64(0) // fileoff
 	g.write64(0) // filesize
-
 	g.write32(0) // maxprot
 	g.write32(0) // initprot
 	g.write32(0) // nsects
 	g.write32(0) // flags
 }
 
+fn (mut g Gen) macho_add_loadcommand(typ u32, size int) {
+	g.macho_ncmds++
+	g.macho_cmdsize += size
+	g.write32(int(typ))
+	g.write32(size)
+}
+
+fn (mut g Gen) macho_patch_header() {
+	g.write32_at(0x10, g.macho_ncmds)
+	g.write32_at(0x14, g.macho_cmdsize)
+}
+
+// probably unnecessary
+fn (mut g Gen) macho_chained_fixups() {
+	g.macho_add_loadcommand(native.lc_dyld_chained_fixups, 16)
+	g.write32(0x4000) // dataoff
+	g.write32(56) // datasize
+
+	g.macho_add_loadcommand(native.lc_dyld_exports_trie, 16)
+	g.write32(0x4000) // dataoff
+	g.write32(56) // datasize
+}
+
 fn (mut g Gen) macho_segment64_linkedit() {
-	g.write32(native.lc_segment_64)
+	g.macho_add_loadcommand(native.lc_segment_64, 0x48)
-	g.write32(0x48) // cmdsize
 	g.write_string_with_padding('__LINKEDIT', 16)
 
-	g.write64(0x3000) // vmaddr
+	// g.size_pos << g.buf.len
-	g.write64(0x1000) // vmsize
+	g.write64(native.base_addr + g.get_pagesize()) // vmaddr
-	g.write64(0x1000) // fileoff
+	g.write64(g.get_pagesize()) // vmsize
+	g.write64(g.get_pagesize()) // fileoff
 	g.write64(0) // filesize
-	g.write32(7) // maxprot
+	g.write32(1) // maxprot
-	g.write32(3) // initprot
+	g.write32(1) // initprot
 	g.write32(0) // nsects
 	g.write32(0) // flags
 }
@@ -85,40 +110,50 @@ fn (mut g Gen) macho_header(ncmds int, bintype int) int {
 	cmdsize_offset := g.buf.len
 	g.write32(0) // size of load commands
 
+	if g.pref.arch == .arm64 {
+		g.write32(0x00200085)
+	} else {
 		g.write32(0) // flags
+	}
 	g.write32(0) // reserved
 	return cmdsize_offset
 }
 
 fn (mut g Gen) macho_segment64_text() []int {
 	mut patch := []int{}
-	g.write32(native.lc_segment_64) // LC_SEGMENT_64
+	g.macho_add_loadcommand(native.lc_segment_64, 152)
-	g.write32(152) // 152
 	g.write_string_with_padding('__TEXT', 16) // section name
-	g.write64(0x100000000) // vmaddr
+	g.write64(native.base_addr) // vmaddr
-	patch << g.buf.len
-	g.write64(0x00001000) // + codesize) // vmsize
-	g.write64(0x00000000) // filesize
-	patch << g.buf.len
-	g.write64(0x00001000) // + codesize) // filesize
 
-	g.write32(7) // maxprot
+	g.write64(g.get_pagesize() * 2) // vmsize
+	g.write64(0) // fileoff
+	g.write64(g.get_pagesize() + 63) // filesize
+
+	g.write32(5) // maxprot
 	g.write32(5) // initprot
 	g.write32(1) // nsects
 	g.write32(0) // flags
 
 	g.write_string_with_padding('__text', 16) // section name
 	g.write_string_with_padding('__TEXT', 16) // segment name
-	g.write64(0x0000000100001000) // vmaddr
+	g.write64(native.base_addr + g.get_pagesize()) // vmaddr
+	if g.pref.arch == .arm64 {
+		g.write64(0) // vmsize
+	} else {
 		patch << g.buf.len
 		g.write64(0) // vmsize
-	g.write32(4096) // offset
+	}
-	g.write32(0) // align
+	g.write32(0) // offset
+	g.write32(4) // align
 
 	g.write32(0) // reloff
 	g.write32(0) // nreloc
 
+	if g.pref.arch == .amd64 {
 		g.write32(0) // flags
+	} else {
+		g.write32(0x80000400) // flags
+	}
 	g.write32(0)
 
 	g.write32(0) // reserved1
@@ -127,16 +162,25 @@ fn (mut g Gen) macho_segment64_text() []int {
 }
 
 fn (mut g Gen) macho_symtab() {
-	g.write32(native.lc_symtab)
+	g.macho_add_loadcommand(native.lc_dyld_info_only, 48)
-	g.write32(24)
+	g.write32(0) // rebase_off
-	g.write32(0x1000)
+	g.write32(0) // rebase_size
-	g.write32(0)
+	g.write32(0) // bind_off
-	g.write32(0x1000)
+	g.write32(0) // bind_size
-	g.write32(0)
+	g.write32(0) // weak_bind_off
+	g.write32(0) // weak_bind_size
+	g.write32(0) // lazy_bind_off
+	g.write32(0) // lazy_bind_size
+	g.write32(0x4000) // export_off
+	g.write32(56) // export_size
 
-	// lc_dysymtab
+	g.macho_add_loadcommand(native.lc_symtab, 24)
-	g.write32(native.lc_dysymtab)
+	g.write32(0x1000) // symoff
-	g.write32(0x50)
+	g.write32(0) // nsyms
+	g.write32(0x1000) // stroff
+	g.write32(0) // strsize
+
+	g.macho_add_loadcommand(native.lc_dysymtab, 0x50)
 	g.write32(0) // ilocalsym
 	g.write32(0) // nlocalsym
 	g.write32(0) // iextdefsym
@@ -158,47 +202,54 @@ fn (mut g Gen) macho_symtab() {
 }
 
 fn (mut g Gen) macho_dylibs() {
-	g.write32(native.lc_load_dylinker)
+	g.macho_add_loadcommand(native.lc_load_dylinker, 32)
-	g.write32(32) // cmdsize (must be aligned to int32)
 	g.write32(12) // offset
 	g.write_string_with_padding('/usr/lib/dyld', 16)
 	g.write32(0) // padding // can be removed
 
-	g.write32(native.lc_load_dylib)
+	g.macho_add_loadcommand(native.lc_load_dylib, 56)
-	g.write32(56) // cmdsize
 	g.write32(24) // offset
 	g.write32(0) // ts
-	g.write32(1) // ver
+	g.write32(0x051f6403) // g.write32(1) // current version
-	g.write32(1) // compat
+	g.write32(0x10000) // compatibility version
 	g.write_string_with_padding('/usr/lib/libSystem.B.dylib', 32)
 }
 
 fn (mut g Gen) macho_main(addr int) {
-	g.write32(int(native.lc_main)) // LC_MAIN
+	g.macho_add_loadcommand(native.lc_main, 24)
-	g.write32(24) // cmdsize
 	g.write32(addr) // entrypoint
 	g.write32(0) // initial_stacksize
 }
 
 pub fn (mut g Gen) generate_macho_header() {
-	g.code_start_pos = 0x1000
+	pagesize := g.get_pagesize()
-	g.debug_pos = 0x1000
+	g.code_start_pos = pagesize
-	cmdsize_offset := g.macho_header(8, native.mh_execute)
+	g.debug_pos = pagesize
+	ncmds := 0 // 9+ 2 -2 -3 -1
+	cmdsize_offset := g.macho_header(ncmds, native.mh_execute)
 	g.macho_segment64_pagezero()
 
 	g.size_pos = g.macho_segment64_text()
-	g.macho_segment64_linkedit()
+	// g.macho_segment64_linkedit()
+	// g.macho_chained_fixups()
 	g.macho_symtab()
 	g.macho_dylibs()
-	g.macho_main(0x1000)
+	g.macho_main(pagesize)
 
 	g.write32_at(cmdsize_offset, g.buf.len - 24)
-	g.write_nulls(0x1000 - g.buf.len)
+	g.write_nulls(pagesize - g.buf.len)
 	g.call(0)
 }
 
+fn (mut g Gen) get_pagesize() int {
+	if g.pref.arch == .arm64 {
+		return 0x4000 // 16KB
+	}
+	return 0x1000 // 4KB
+}
+
 fn (mut g Gen) write_nulls(len int) {
-	pad := 0x1000 - g.buf.len
+	pad := g.get_pagesize() - g.buf.len
 	for _ in 0 .. pad {
 		g.write8(0)
 	}
@@ -208,11 +259,11 @@ pub fn (mut g Gen) generate_macho_object_header() {
 	if g.pref.arch == .arm64 {
 		g.write32(0xfeedfacf) // MH_MAGIC_64
 		g.write32(0x0100000c) // CPU_TYPE_ARM64
-		g.write32(0x00000000) // CPU_SUBTYPE_ARM64_ALL
+		g.write32(0) // CPU_SUBTYPE_ARM64_ALL
 	} else {
 		g.write32(0xfeedfacf) // MH_MAGIC_64
 		g.write32(0x01000007) // CPU_TYPE_X64
-		g.write32(0x00000003) // CPU_SUBTYPE_X64
+		g.write32(3) // CPU_SUBTYPE_X64
 	}
 	g.write32(native.mh_object) // MH_OBJECT
 	text_offset := 0x138
@@ -249,7 +300,6 @@ pub fn (mut g Gen) generate_macho_object_header() {
 	/// ???
 	g.write32(0x32)
 	g.write32(0x18)
-
 	g.write32(0x01)
 	g.write32(0x000a0000) // minOS 10.0
 	g.write32(0)
@@ -257,8 +307,7 @@ pub fn (mut g Gen) generate_macho_object_header() {
 	// lc_symtab
 	g.sym_table_command()
 	//
-	g.write32(native.lc_dysymtab)
+	g.macho_add_loadcommand(native.lc_dysymtab, native.macho_d_size)
-	g.write32(native.macho_d_size)
 	g.write32(0)
 	g.write32(2)
 	g.write32(2)
@@ -291,13 +340,26 @@ pub fn (mut g Gen) generate_macho_footer() {
 	g.write8(0)
 	for o in g.size_pos {
 		n := g.read32_at(o)
+		// eprintln('$n + $delta')
 		g.write32_at(o, n + delta)
 	}
 	g.write64(0)
-	// this is amd64-specific
+	g.macho_patch_header()
+	if g.pref.arch == .amd64 {
 		call_delta := int(g.main_fn_addr - g.code_start_pos) - 5
 		g.write32_at(g.code_start_pos + 1, call_delta)
 		g.create_executable()
+	} else {
+		call_delta := int(g.main_fn_addr - g.code_start_pos)
+		if (call_delta % 4) != 0 || call_delta < 0 {
+			g.n_error('Invalid entrypoint->main delta ($call_delta)')
+		} else {
+			blop := (0x94 << 24) | (call_delta / 4)
+			g.write32_at(g.code_start_pos, int(blop))
+			g.write_nulls(g.get_pagesize() - g.buf.len)
+			g.create_executable()
+		}
+	}
 }
 
 fn (mut g Gen) sym_table_command() {
@@ -338,8 +400,7 @@ fn (mut g Gen) sym_table_command() {
 		name: 'ltmp1'
 		is_ext: false
 	}
-	g.write32(native.lc_symtab)
+	g.macho_add_loadcommand(native.lc_symtab, native.macho_symcmd_size)
-	g.write32(native.macho_symcmd_size)
 	sym_table_offset := 0x168
 	g.write32(sym_table_offset)
 	g_syms_len := 4
@@ -406,8 +467,7 @@ fn (mut g Gen) sym_string_table() int {
 					// that should be .rel32, not windows-specific
 					g.write32_at(s.pos, pos)
 				} else {
-					baddr := i64(0x100000000)
+					g.write64_at(s.pos, g.buf.len + native.base_addr)
-					g.write64_at(s.pos, g.buf.len + baddr)
 				}
 			}
 		}

vlib/v/gen/native/macho_test.v

@@ -6,7 +6,9 @@ import v.ast
 fn test_macho() {
 	os.chdir(os.temp_dir()) or {}
 	mut g := native.Gen{
-		pref: &pref.Preferences{}
+		pref: &pref.Preferences{
+			arch: .amd64
+		}
 		out_name: 'test.bin'
 		table: ast.new_table()
 		code_gen: native.Amd64{