From 987c71d9bcf5f1211dfde6e515d8a35675ca5983 Mon Sep 17 00:00:00 2001 From: Richard Kenner Date: Mon, 21 Mar 1994 17:14:41 -0500 Subject: [PATCH] (expand_expr, case ADDR_EXPR): Update temp slot address if we put it in a register; mark result as pointer. From-SVN: r6842 --- gcc/expr.c | 19 ++++++++++++++++++- 1 file changed, 18 insertions(+), 1 deletion(-) diff --git a/gcc/expr.c b/gcc/expr.c index 130d40666ad..7a49abfa3d7 100644 --- a/gcc/expr.c +++ b/gcc/expr.c @@ -5779,6 +5779,10 @@ expand_expr (exp, target, tmode, modifier) return expand_increment (exp, ! ignore); case ADDR_EXPR: + /* If nonzero, TEMP will be set to the address of something that might + be a MEM corresponding to a stack slot. */ + temp = 0; + /* Are we taking the address of a nested function? */ if (TREE_CODE (TREE_OPERAND (exp, 0)) == FUNCTION_DECL && decl_function_context (TREE_OPERAND (exp, 0)) != 0) @@ -5803,6 +5807,8 @@ expand_expr (exp, target, tmode, modifier) if (CONSTANT_P (op0)) op0 = force_const_mem (TYPE_MODE (TREE_TYPE (TREE_OPERAND (exp, 0))), op0); + else if (GET_CODE (op0) == MEM) + temp = XEXP (op0, 0); /* These cases happen in Fortran. Is that legitimate? Should Fortran work in another way? @@ -5827,10 +5833,21 @@ expand_expr (exp, target, tmode, modifier) if (modifier == EXPAND_SUM || modifier == EXPAND_INITIALIZER) return XEXP (op0, 0); + op0 = force_operand (XEXP (op0, 0), target); } + if (flag_force_addr && GET_CODE (op0) != REG) - return force_reg (Pmode, op0); + op0 = force_reg (Pmode, op0); + + if (GET_CODE (op0) == REG) + mark_reg_pointer (op0); + + /* If we might have had a temp slot, add an equivalent address + for it. */ + if (temp != 0) + update_temp_slot_address (temp, op0); + return op0; case ENTRY_VALUE_EXPR: -- 2.30.2