[PATCH] D28308: [SCEV] Model ashr(shl(x, n), m), where n > m, as mul(x, 2^(n-m))

[Eli Friedman via Phabricator via llvm-commits]

efriedma added a comment.

Please upload patches with full context (`-U1000000`).



================
Comment at: lib/Analysis/ScalarEvolution.cpp:5276
                 BO->LHS->getType());
+          } else if (ConstantInt *LCI = dyn_cast<ConstantInt>(LOp1)) {
+            // %y = shl %x, n
----------------
Maybe reorganize the code to look more like this (rough outline)?

    ConstantInt *LCI = dyn_cast<ConstantInt>(LOp1));
    if (!LCI) break;
    if (ashr greater than shl) break;
    if (ashr less than than shl) {
      // Multiply SCEV
    }
    return sext(trunc(scev));


================
Comment at: test/Analysis/ScalarEvolution/sext-mul.ll:4
+; CHECK:  %12 = ashr exact i64 %11, 32
+; CHECK:  -->  {0,+,2}<%9> U: [0,-1) S: [-9223372036854775808,9223372036854775807)              Exits: (-2 + (2 * (zext i32 %2 to i64)))<nsw>           LoopDispositions: { %9: Computable }
+; CHECK:  %13 = getelementptr inbounds i32, i32* %0, i64 %12
----------------
The loop induction variable (let's call it `i`) translates to `{0,+,1}<%9>`.  `i*2` translates to `{0,+,2}<%9>`.  `sext(trunc(i to i32) to i64)` `(sext i32 {0,+,1}<%9> to i64)`.  `sext(trunc(i*2 to i32) to i64)` translates to `(sext i32 {0,+,2}<%9> to i64)`.  `ashr(shl(i, 33), 32)` should also translate to `(sext i32 {0,+,2}<%9> to i64)`.

With your patch, `ashr(shl(i, 33), 32)` translates to `{0,+,2}<%9>`, which is wrong because the loop can iterate more that INT_MAX times.


Repository:
  rL LLVM

https://reviews.llvm.org/D28308