[PATCH] D75449: [MLIR][GPU] fix loop trip count computation in LoopsToGPU

[Tobias Gysi via Phabricator via llvm-commits]

gysit created this revision.
gysit added a reviewer: herhut.
Herald added subscribers: llvm-commits, Joonsoo, liufengdb, lucyrfox, mgester, arpith-jacob, csigg, nicolasvasilache, antiagainst, shauheen, burmako, jpienaar, rriddle, mehdi_amini.
Herald added a reviewer: mravishankar.
Herald added a project: LLVM.

Added brackets to fix the loop trip count computation. The brackets ensure the bounds are subtracted before we divide the result by the step of the loop.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D75449

Files:
  mlir/lib/Conversion/LoopsToGPU/LoopsToGPU.cpp
  mlir/test/Conversion/LoopsToGPU/parallel_loop.mlir


Index: mlir/test/Conversion/LoopsToGPU/parallel_loop.mlir
===================================================================
--- mlir/test/Conversion/LoopsToGPU/parallel_loop.mlir
+++ mlir/test/Conversion/LoopsToGPU/parallel_loop.mlir
@@ -15,7 +15,7 @@
   return
 }
 
-// CHECK:       #[[MAP0:.*]] = affine_map<()[s0, s1, s2] -> (s0 - s1 ceildiv s2)>
+// CHECK:       #[[MAP0:.*]] = affine_map<()[s0, s1, s2] -> ((s0 - s1) ceildiv s2)>
 // CHECK:       #[[MAP1:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s0 + s1)>
 
 // CHECK:       module {
@@ -66,7 +66,7 @@
   return
 }
 
-// CHECK:       #[[MAP0:.*]] = affine_map<()[s0, s1, s2] -> (s0 - s1 ceildiv s2)>
+// CHECK:       #[[MAP0:.*]] = affine_map<()[s0, s1, s2] -> ((s0 - s1) ceildiv s2)>
 // CHECK:       #[[MAP1:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s0 + s1)>
 
 // CHECK:       module {
@@ -115,7 +115,7 @@
   return
 }
 
-// CHECK:       #[[MAP0:.*]] = affine_map<()[s0, s1, s2] -> (s0 - s1 ceildiv s2)>
+// CHECK:       #[[MAP0:.*]] = affine_map<()[s0, s1, s2] -> ((s0 - s1) ceildiv s2)>
 // CHECK:       #[[MAP1:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s0 + s1)>
 
 // CHECK:       module {
@@ -166,7 +166,7 @@
   return
 }
 
-// CHECK:       #[[MAP0:.*]] = affine_map<()[s0, s1, s2] -> (s0 - s1 ceildiv s2)>
+// CHECK:       #[[MAP0:.*]] = affine_map<()[s0, s1, s2] -> ((s0 - s1) ceildiv s2)>
 // CHECK:       #[[MAP1:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s0 + s1)>
 
 // CHECK:       module {
@@ -241,7 +241,7 @@
 }
 
 // CHECK:       #[[MAP0:.*]] = affine_map<(d0, d1)[s0, s1] -> (d0 * s1 + s0 + d1)>
-// CHECK:       #[[MAP1:.*]] = affine_map<()[s0, s1, s2] -> (s0 - s1 ceildiv s2)>
+// CHECK:       #[[MAP1:.*]] = affine_map<()[s0, s1, s2] -> ((s0 - s1) ceildiv s2)>
 // CHECK:       #[[MAP2:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s0 + s1)>
 // CHECK:       #[[MAP3:.*]] = affine_map<(d0)[s0] -> (2, -d0 + s0)>
 // CHECK:       #[[MAP4:.*]] = affine_map<(d0)[s0] -> (3, -d0 + s0)>
Index: mlir/lib/Conversion/LoopsToGPU/LoopsToGPU.cpp
===================================================================
--- mlir/lib/Conversion/LoopsToGPU/LoopsToGPU.cpp
+++ mlir/lib/Conversion/LoopsToGPU/LoopsToGPU.cpp
@@ -653,8 +653,8 @@
           // affine.apply here so that it composes nicely with the provided map.
           AffineMap stepMap =
               AffineMap::get(0, 3,
-                             (rewriter.getAffineSymbolExpr(0) -
-                              rewriter.getAffineSymbolExpr(1).ceilDiv(
+                             ((rewriter.getAffineSymbolExpr(0) -
+                              rewriter.getAffineSymbolExpr(1)).ceilDiv(
                                   rewriter.getAffineSymbolExpr(2))));
           Value launchBound = rewriter.create<AffineApplyOp>(
               loc, annotation.boundMap.compose(stepMap),


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D75449.247613.patch
Type: text/x-patch
Size: 2806 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200302/81c4054f/attachment.bin>