[PATCH] D78055: [MLIR] remove dead affine.applys while generating pointwise copies

[Uday Bondhugula via Phabricator via llvm-commits]

bondhugula created this revision.
bondhugula added a reviewer: andydavis1.
Herald added subscribers: llvm-commits, frgossen, grosul1, Joonsoo, liufengdb, lucyrfox, mgester, arpith-jacob, nicolasvasilache, antiagainst, shauheen, burmako, jpienaar, rriddle, mehdi_amini.
Herald added a project: LLVM.

This makes no impact on the test cases because affine-data-copy-generate
runs whole function canonicalization at its end; however, the latter
will be removed in a pending revision. It is thus useful to clean up
these affine.applys right here, and eventually, not even generate
these (when the right API to compose by construction is in place).


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D78055

Files:
  mlir/lib/Transforms/Utils/LoopUtils.cpp


Index: mlir/lib/Transforms/Utils/LoopUtils.cpp
===================================================================
--- mlir/lib/Transforms/Utils/LoopUtils.cpp
+++ mlir/lib/Transforms/Utils/LoopUtils.cpp
@@ -1492,6 +1492,7 @@
   SmallVector<AffineExpr, 4> fastBufExprs;
   SmallVector<Value, 4> fastBufMapOperands;
   AffineForOp copyNestRoot;
+  SmallVector<AffineApplyOp, 4> mayBeDeadApplys;
   for (unsigned d = 0; d < rank; ++d) {
     auto forOp = createCanonicalizedAffineForOp(b, loc, lbOperands, lbMaps[d],
                                                 ubOperands, ubMaps[d]);
@@ -1510,6 +1511,7 @@
                            b.getAffineDimExpr(2 * d));
     fastBufMapOperands.push_back(offset);
     fastBufMapOperands.push_back(forOp.getInductionVar());
+    mayBeDeadApplys.push_back(offset);
 
     // Subscript for the slow memref being copied.
     memIndices.push_back(forOp.getInductionVar());
@@ -1520,6 +1522,11 @@
   fastBufMap = simplifyAffineMap(fastBufMap);
   canonicalizeMapAndOperands(&fastBufMap, &fastBufMapOperands);
 
+  // Drop any dead affine.applys.
+  for (auto applyOp : mayBeDeadApplys)
+    if (applyOp.use_empty())
+      applyOp.erase();
+
   if (!isCopyOut) {
     // Copy in.
     auto load = b.create<AffineLoadOp>(loc, memref, memIndices);
@@ -2191,7 +2198,7 @@
   // larger (and resp. smaller) than any other lower (or upper bound).
   SmallVector<int64_t, 8> fullTileLb, fullTileUb;
   for (auto loop : loops) {
-    (void) loop;
+    (void)loop;
     // TODO: Non-unit stride is not an issue to generalize to.
     assert(loop.getStep() == 1 && "point loop step expected to be one");
     // Mark everything symbols for the purpose of finding a constant diff pair.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D78055.257108.patch
Type: text/x-patch
Size: 1714 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200413/819d9f88/attachment.bin>