Fix ALLOCA to align stack frame

TODO

@@ -66,7 +66,7 @@
 	-  CAST
 	+  TAILCALL
 	+  VLOAD, VSTORE, VADDR
-	?  ALLOCA, LOAD, STORE
+	+  ALLOCA, LOAD, STORE
 	+  SWITCH
 	?  ir_last_use
 	+  binop_int $imm, mem

ir_x86.dasc

@@ -3241,10 +3241,40 @@ static void ir_emit_alloca(ir_ctx *ctx, ir_ref def, ir_insn *insn)
 	ir_backend_data *data = ctx->data;
 	dasm_State **Dst = &data->dasm_state;
 
-	// TODO: alignment
-	|	ASM_REG_REF_OP sub, IR_ADDR, IR_REG_RSP, insn->op2
+	if (IR_IS_CONST_REF(insn->op2)) {
+		ir_insn *val = &ctx->ir_base[insn->op2];
+		int32_t size = val->val.i32;
+
+		IR_ASSERT(IR_IS_TYPE_INT(val->type));
+		IR_ASSERT(IR_IS_TYPE_UNSIGNED(val->type) || val->val.i64 > 0);
+		IR_ASSERT(IR_IS_SIGNED_32BIT(val->val.i64));
+
+		if (ctx->flags & IR_HAS_CALLS) {
+			/* Stack must be 16 byte aligned */
+			size = IR_ALIGNED_SIZE(size, 16);
+		} else {
+			size = IR_ALIGNED_SIZE(size, 8);
+		}
+		|	ASM_REG_IMM_OP sub, IR_ADDR, IR_REG_RSP, size
+	} else {
+		int32_t alignment = (ctx->flags & IR_HAS_CALLS) ? 16 : 8;
+		ir_reg op2_reg = ir_ref_reg(ctx, insn->op2);
+		ir_reg reg;
+
+		if (op2_reg >=0 && ir_last_use(ctx, insn->op2, def)) {
+			reg = op2_reg;
+		} else {
+		    reg = IR_REG_RAX; // TODO: temporary register
+		}
+		if (reg != op2_reg) {
+			ir_emit_load(ctx, IR_ADDR, insn->op2, reg);
+		}
+
+		|	ASM_REG_IMM_OP add, IR_ADDR, reg, (alignment-1)
+		|	ASM_REG_IMM_OP and, IR_ADDR, reg, ~(alignment-1)
+		|	ASM_REG_REG_OP sub, IR_ADDR, IR_REG_RSP, reg
+	}
 	ir_emit_store(ctx, IR_ADDR, IR_REG_RSP, def);
-	// TODO: stack frame alignment
 }
 
 static void ir_emit_switch(ir_ctx *ctx, ir_ref def, ir_insn *insn)

tests/debug/call_alloca.irt

@@ -0,0 +1,67 @@
+--TEST--
+Simple CALL with ALLOCA
+--ARGS--
+-S --run
+--CODE--
+{
+	uintptr_t c_1 = 0;
+	bool c_2 = 0;
+	bool c_3 = 1;
+	int32_t c_4 = 42;
+	char h = 'h';
+	char i = 'i';
+	char sp = ' ';
+	char pc = '%';
+	char d = 'd';
+	char nl = '\n';
+	char zero = 0;
+	uint32_t len = 7;
+	uintptr_t one = 1;
+	uintptr_t c_5 = func(printf);
+	l_1 = START(l_4);
+	uintptr_t a_0, s_0 = ALLOCA(l_1, len);
+	s_1 = STORE(s_0, a_0, h);
+	uintptr_t a_1 = ADD(a_0, one);
+	s_2 = STORE(s_1, a_1, i);
+	uintptr_t a_2 = ADD(a_1, one);
+	s_3 = STORE(s_2, a_2, sp);
+	uintptr_t a_3 = ADD(a_2, one);
+	s_4 = STORE(s_3, a_3, pc);
+	uintptr_t a_4 = ADD(a_3, one);
+	s_5 = STORE(s_4, a_4, d);
+	uintptr_t a_5 = ADD(a_4, one);
+	s_6 = STORE(s_5, a_5, nl);
+	uintptr_t a_6 = ADD(a_5, one);
+	s_7 = STORE(s_6, a_6, zero);
+	int32_t d_2, l_2 = CALL/2(s_7, c_5, a_0, c_4);
+	l_4 = RETURN(l_2, d_2);
+}
+--EXPECT--
+test:
+	pushq %rbp
+	movq %rsp, %rbp
+	subq $0x10, %rsp
+	movq %rsp, %rdi
+	movb $0x68, (%rdi)
+	leaq 1(%rdi), %rax
+	movb $0x69, (%rax)
+	leaq 2(%rdi), %rax
+	movb $0x20, (%rax)
+	leaq 3(%rdi), %rax
+	movb $0x25, (%rax)
+	leaq 4(%rdi), %rax
+	movb $0x64, (%rax)
+	leaq 5(%rdi), %rax
+	movb $0xa, (%rax)
+	leaq 6(%rdi), %rax
+	movb $0, (%rax)
+	movl $0x2a, %esi
+	movabsq $_IO_printf, %rax
+	callq *%rax
+	movq %rbp, %rsp
+	popq %rbp
+	retq
+
+hi 42
+
+exit code = 6

tests/debug/test_mem.irt

@@ -88,7 +88,7 @@ test:
 	movsd %xmm7, -0x28(%rbp)
 	xorpd %xmm7, %xmm7
 	movsd %xmm7, -0x20(%rbp)
-	subq $4, %rsp
+	subq $8, %rsp
 	movq %rsp, %rax
 	movl $0, (%rax)
 .L1: