Re: [PATCH 1/2] sparse, llvm: group PHI nodes at the top of each BB

[Xi Wang <xi.wang@gmail.com>]

On 10/16/12 4:14 PM, Jonathan Neuschäfer wrote:
> Actually, the situation of Phi nodes in LLVM is actually slightly more
> complex: They require "one pair (of value and BB) for each predecessor
> basic block of the current block"[1]. This mean that we'll sometimes
> need to insert phi nodes into BBs that don't directly use a value.

I ran into the same problem before.  I would suggest a simpler and safer
way: don't emit LLVM phi; instead emit load/store (of some alloca).

phi    => load from some alloca
phisrc => store to some alloca

You can find sample code from splay here (emit_phi & emit_phisrc):

https://github.com/xiw/splay/blob/master/function.c#L356

> Consider the following piece of C code:
> 
> 	extern int done(void);
> 	extern void foo(int);
> 
> 	static void test(void) {
> 		int i;
> 		for (i = 0; ; i++) {
> 			if (done())
> 				break;
> 			foo(i);
> 		}
> 	}

This is what splay emits:

define internal void @test() {
entry:
  %0 = alloca i32
  store i32 0, i32* %0
  br label %bb

bb:                                               ; preds = %bb1, %entry
  %1 = call i32 @done()
  %2 = icmp ne i32 %1, 0
  br i1 %2, label %bb2, label %bb1

bb1:                                              ; preds = %bb
  %3 = load i32* %0
  call void @foo(i32 %3)
  %4 = add nsw i32 %3, 1
  store i32 %4, i32* %0
  br label %bb

bb2:                                              ; preds = %bb
  ret void
}

After "-mem2reg" you get:

define internal void @test() {
entry:
  br label %bb

bb:                                               ; preds = %bb1, %entry
  %.0 = phi i32 [ 0, %entry ], [ %2, %bb1 ]
  %0 = call i32 @done()
  %1 = icmp ne i32 %0, 0
  br i1 %1, label %bb2, label %bb1

bb1:                                              ; preds = %bb
  call void @foo(i32 %.0)
  %2 = add nsw i32 %.0, 1
  br label %bb

bb2:                                              ; preds = %bb
  ret void
}

- xi
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html