cgen,parser: fix arm64 asm operand position; fix arm64 asm imm; support arm64 dot instruction (#24017)

vlib/v/gen/c/cgen.v

@@ -3130,8 +3130,8 @@ fn (mut g Gen) asm_stmt(stmt ast.AsmStmt) {
 		} else {
 			g.write(' ')
 		}
-		// swap destination and operands for att syntax
+		// swap destination and operands for att syntax, not for arm64
-		if template.args.len != 0 && !template.is_directive {
+		if template.args.len != 0 && !template.is_directive && stmt.arch != .arm64 {
 			template.args.prepend(template.args.last())
 			template.args.delete(template.args.len - 1)
 		}
@@ -3206,8 +3206,12 @@ fn (mut g Gen) asm_arg(arg ast.AsmArg, stmt ast.AsmStmt) {
 			g.write("'${arg.val}'")
 		}
 		ast.IntegerLiteral {
+			if stmt.arch == .arm64 {
+				g.write('#${arg.val}')
+			} else {
 				g.write('\$${arg.val}')
 			}
+		}
 		ast.FloatLiteral {
 			if g.pref.nofloat {
 				g.write('\$${arg.val.int()}')

vlib/v/parser/parser.v

@@ -1266,6 +1266,7 @@ fn (mut p Parser) asm_stmt(is_top_level bool) ast.AsmStmt {
 	for p.tok.kind !in [.semicolon, .rcbr, .eof] {
 		template_pos := p.tok.pos()
 		mut name := ''
+		mut comments := []ast.Comment{}
 		if p.tok.kind == .name && arch == .amd64 && p.tok.lit in ['rex', 'vex', 'xop'] {
 			name += p.tok.lit
 			p.next()
@@ -1297,12 +1298,16 @@ fn (mut p Parser) asm_stmt(is_top_level bool) ast.AsmStmt {
 		} else if p.tok.kind == .number {
 			name += p.tok.lit
 			p.next()
+		} else if p.tok.kind == .comment {
+			for p.tok.kind == .comment {
+				comments << p.comment()
+			}
 		} else {
 			name += p.tok.lit
 			p.check(.name)
 		}
-		// dots are part of instructions for some riscv extensions and webassembly
+		// dots are part of instructions for some riscv extensions and webassembly, arm64
-		if arch in [.rv32, .rv64, .wasm32] {
+		if arch in [.rv32, .rv64, .wasm32, .arm64] {
 			for p.tok.kind == .dot {
 				name += '.'
 				p.next()
@@ -1405,7 +1410,6 @@ fn (mut p Parser) asm_stmt(is_top_level bool) ast.AsmStmt {
 			// 	break
 			// }
 		}
-		mut comments := []ast.Comment{}
 		for p.tok.kind == .comment {
 			comments << p.comment()
 		}

vlib/v/slow_tests/assembly/asm_test.arm64.v

@@ -0,0 +1,124 @@
+// vtest build: gcc
+fn test_inline_asm() {
+	a, mut b := 10, 0
+	asm arm64 {
+		mov x0, a
+		mov b, x0
+		; +r (b)
+		; r (a)
+		; x0
+	}
+	assert a == 10
+	assert b == 10
+
+	mut c := 0
+	asm arm64 {
+		mov x0, 5
+		mov c, x0
+		; +r (c)
+		; ; x0
+	}
+	assert c == 5
+
+	d, e, mut f := 10, 2, 0
+	asm arm64 {
+		mov x0, d
+		mov x1, e
+		add x0, x0, x1
+		add x0, x0, 5
+		mov f, x0
+		; +r (f) // output
+		; r (d)
+		  r (e) // input
+		; x0
+		  x1
+	}
+	assert d == 10
+	assert e == 2
+	assert f == 17
+
+	mut j := 0
+	// do 5*3
+	// adding three, five times
+	asm arm64 {
+		mov x0, 5 // loop 5 times
+		mov x1, 0
+		loop_start:
+		add x1, x1, 3
+		sub x0, x0, 1
+		cmp x0, 0
+		b.gt loop_start
+		mov j, x1
+		; +r (j)
+		; ; x0
+		  x1
+	}
+	assert j == 5 * 3
+	/*
+	// not marked as mut because we dereference m to change l
+	l := 5
+	m := &l
+	asm arm64 {
+		movd [m], 7 // have to specify size with q
+		; ; r (m)
+	}
+	assert l == 7
+
+	mut manu := Manu{}
+	asm arm64 {
+		mov x0, MIDR_EL1
+		mov x1, ID_AA64ISAR0_EL1
+		mov x2, ID_AA64MMFR0_EL1
+		; =r (manu.midr_el1) as x0
+		  =r (manu.id_aa64isar0_el1) as x1
+		  =r (manu.id_aa64mmfr0_el1) as x2
+	}
+	manu.str()
+	*/
+}
+
+/*
+@[packed]
+struct Manu {
+mut:
+	midr_el1  u64
+	id_aa64isar0_el1  u64
+	id_aa64mmfr0_el1  u64
+	zero u8 // for string
+}
+
+fn (m Manu) str() string {
+	return unsafe {
+		string{
+			str:    &u8(&m)
+			len:    24
+			is_lit: 1
+		}
+	}
+}
+
+fn test_asm_generic() {
+	u := u64(49)
+	b := unsafe { bool(123) }
+	assert generic_asm(u) == 14
+	assert u == 63
+	assert u64(generic_asm(b)) == 14
+	assert u64(b) == 137
+}
+
+fn generic_asm[T](var &T) T {
+	mut ret := T(14)
+	unsafe {
+		asm volatile arm64 {
+			mov x0, ret
+			mov x1, var
+			add x1, x0, 0
+			mov var, x1
+			; +m (var[0]) as var
+			  +r (ret)
+			; ; memory x0 x1
+		}
+	}
+	return ret
+}
+*/