sparse, llvm: improve pointer arithmetic handling
PATCH 1/2 sparse, llvm: improve pointer arithmetic handling
Converting pointers to integers for pointer arithmetic effectively disables pointer analysis and future optimizations. A better way is to use LLVM's GEP, by converting pointers to `char *' rather than integers.
---calc_gep() may be useful for fixing array access as well.
--- sparse-llvm.c | 38 ++++++++++++++++++++++++++------------ 1 file changed, 26 insertions(+), 12 deletions(-)
sparse-llvm.c +19 -9
@@ -370,6 +370,22 @@ static LLVMValueRef pseudo_to_value(struct function *fn, struct instruction *insreturn result; } +static LLVMValueRef calc_gep (LLVMBuilderRef builder,LLVMValueRef base,LLVMValueRef off) +{ +LLVMTypeRef type =LLVMTypeOf (base); +unsigned int as =LLVMGetPointerAddressSpace (type); +LLVMTypeRef bytep =LLVMPointerType (LLVMInt8Type (), as); +LLVMValueRef addr; + +/* convert base to char* type */ + base =LLVMBuildPointerCast (builder, base, bytep,"" ); +/* addr = base + off */ + addr =LLVMBuildInBoundsGEP (builder, base, &off,1 ,"" ); +/* convert back to the actual pointer type */ + addr =LLVMBuildPointerCast (builder, addr, type,"" ); +return addr; +} +static LLVMRealPredicate translate_fop (int opcode) {static const LLVMRealPredicate trans_tbl[] = {
@@ -544,23 +560,21 @@ static void output_op_ret(struct function *fn, struct instruction *insn)static LLVMValueRef calc_memop_addr (struct function *fn,struct instruction *insn) {LLVMTypeRef int_type, addr_type; -LLVMValueRef src_p, src_i, ofs_i, addr_i, addr; +LLVMValueRef src, off, addr; +unsigned int as;/* int type large enough to hold a pointer */ int_type =LLVMIntType (bits_in_pointer); + off =LLVMConstInt (int_type, insn->offset,0 ); -/* convert to integer, add src + offset */ - src_p =pseudo_to_value (fn, insn, insn->src); - src_i =LLVMBuildPtrToInt (fn->builder, src_p, int_type,"src_i" ); - - ofs_i =LLVMConstInt (int_type, insn->offset,0 ); - addr_i =LLVMBuildAdd (fn->builder, src_i, ofs_i,"addr_i" ); - - addr_type =LLVMPointerType (insn_symbol_type (fn->module, insn),0 );
PATCH 2/2 sparse, llvm: set target specification
Set target triple and data layout, which are required by LLVM's backend.
Also export arch_m64 for choosing the target architecture.
---
With the two patches, we will be able to recover struct access.
Consider t.c:
struct A { int x, y; };
void foo(struct A *a)
{
a->x = 123;
a->y = 456;
}$ ./sparse-llvm t.c | opt -S -O2
...
define void @foo(%A* nocapture) #0 {
L0:
%1 = getelementptr inbounds %A* %0, i64 0, i32 0
store i32 123, i32* %1, align 4
%2 = getelementptr inbounds %A* %0, i64 0, i32 1
store i32 456, i32* %2, align 4
ret void
}
---
lib.h | 2 ++
sparse-llvm.c | 56 +++++++++++++++++++++++++++++++++++++++++++++++++++++---
2 files changed, 55 insertions(+), 3 deletions(-)lib.h +2 -0
@@ -113,6 +113,8 @@ extern int Wvla;extern int dbg_entry;extern int dbg_dead; +extern int arch_m64; +extern void declare_builtin_functions (void );extern void create_builtin_stream (void );extern struct symbol_list *sparse_initialize (int argc,char **argv,struct string_list **files);
sparse-llvm.c +53 -3
73 lines not on this page · raw message
Signature · 6 lines
1.8.1.2 -- To unsubscribe from this list: send the line "unsubscribe linux-sparse" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
Quoted 7 lines · Xi Wang
Converting pointers to integers for pointer arithmetic effectively disables pointer analysis and future optimizations. A better way is to use LLVM's GEP, by converting pointers to `char *' rather than integers.
Cc Pekka Enberg <[email protected]>Signed-off-by Xi Wang <[email protected]>
Applied both patches, thanks a lot Xi!
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html