zahed khurasani <sdzahed@xxxxxxxxx> writes: > This is the error I am getting: > > "test8.c:11:7: internal compiler error: in expand_expr_real_1, at expr.c:8755" > > Looking at the code I see that this is an assert statement. The call > to the function > get_addr_base_and_unit_offset() returns NULL, which is expected as per > the comments > on this function. If you use POINTER_PLUS_EXPR, don't also use ADDR_EXPR. POINTER_PLUS_EXPR already calculates an address; you don't want to take the address of that, as there is nothing to take the address of. Anyhow, that assert appears to the case of a MEM_REF of an ADDR_EXPR. That really only makes sense when taking the address of some variable. It doesn't make sense to take a MEM_REF of some variable location; that variable should be expressed in the MEM_REF, not the ADDR_EXPR. Ian > On Fri, Mar 9, 2012 at 9:02 PM, Ian Lance Taylor <iant@xxxxxxxxxx> wrote: >> zahed khurasani <sdzahed@xxxxxxxxx> writes: >> >>> Let me try again. If I have an expression of the from a[i][10], then I >>> need to insert the >>> following function call into the tree: >>> check_deref(a[i][10], sizeof(a[i][10]), &a[i][10]); >>> >>> I tried POINTER_PLUS_EXPR. It fails at the same place. However, it >>> works if I have >>> constant indices like a[1][10]. >>> >>> Hope this is clearer than before. Thanks! >> >> Exactly how does it fail if you have non-constant indices? >> >> Ian >> >>> On Fri, Mar 9, 2012 at 7:48 PM, Ian Lance Taylor <iant@xxxxxxxxxx> wrote: >>>> zahed khurasani <sdzahed@xxxxxxxxx> writes: >>>> >>>>> I need to build a node that is equivalent to the address of the memory >>>>> being derenferenced. >>>>> MEM_REF would help with the first parameter. But what about the third parameter? >>>>> I would need an ADDR_EXPR for this parameter. However, the code I am using >>>>> is failing. >>>>> Thanks for the help. >>>> >>>> I'm not sure I completely understand, but I will say that a >>>> POINTER_PLUS_EXPR would be the address equivalent of a MEM_REF. >>>> >>>> Ian >>>> >>>>> On Fri, Mar 9, 2012 at 6:04 PM, Ian Lance Taylor <iant@xxxxxxxxxx> wrote: >>>>>> zahed khurasani <sdzahed@xxxxxxxxx> writes: >>>>>> >>>>>>> My check function accepts three parameters: the value being >>>>>>> dereferenced, the size of the object and the address in memory. I am >>>>>>> using the following two lines to get the third address parameter. >>>>>>> >>>>>>> addr = build1 (ADDR_EXPR, build_pointer_type (type), t); // t is the >>>>>>> ARRAY_REF node >>>>>>> fold_convert_loc(location, ptr_type_node, addr); // type case to void * >>>>>>> >>>>>>> However, this is failing whenever the array_ref contains a >>>>>>> non-constant expression, say like a[i+1]. The specific failure is >>>>>>> inside the expand_expr_real_1 method in the file expr.c. Looks like >>>>>>> such non-constant indices are not allowed at this stage in the >>>>>>> compilation (I've inserted the pass just before lowering to RTL >>>>>>> happens). >>>>>>> >>>>>>> My question is, what would be the right way to get the address? I am >>>>>>> looking for something that will preferably work for all dereferences >>>>>>> like MEM_REF, INDIRECT_REF etc. >>>>>> >>>>>> ARRAY_REF certainly permits non-constant indices, so I don't think that >>>>>> is your problem. >>>>>> >>>>>> However, MEM_REF is the most general memory reference expression. >>>>>> >>>>>> Ian
