Fixup that scf.parallel induction variables were iterated in wrong (reversed) order, leading to issues in scf.parallel unrolling

Author: erwei-xilinx

mlir/lib/Util/Dependency.cpp

@@ -774,6 +774,10 @@ LogicalResult unrollScfParallel(
     IRMapping localRemap = remap;
     std::vector<unsigned> position =
         air::getMDVectorFromIterator(par_size, iter);
+    std::reverse(
+        position.begin(),
+        position
+            .end()); // scf.parallel induction vars. have LSD at highest index.
     for (unsigned i = 0; i < position.size(); i++) {
       localRemap.map(par.getInductionVars()[i],
                      builder.create<arith::ConstantIndexOp>(

mlir/test/Transform/AIRDependencyScheduleOpt/loop_fusion.mlir

@@ -751,10 +751,10 @@ func.func @func8(%arg0: memref<512x1024xbf16>, %arg1: memref<256x4x4x128xbf16>,
 // CHECK-NEXT: air.channel.get async [{{.*}}]  @channel_1[]
 // CHECK-NEXT: air.channel.put async [{{.*}}]  @channel_2[%c0{{.*}}, %c0{{.*}}]
 // CHECK-NEXT: air.channel.put async [{{.*}}]  @channel_2[%c0{{.*}}, %c0{{.*}}]
-// CHECK-NEXT: air.channel.put async [{{.*}}]  @channel_2[%c1{{.*}}, %c0{{.*}}]
-// CHECK-NEXT: air.channel.put async [{{.*}}]  @channel_2[%c1{{.*}}, %c0{{.*}}]
 // CHECK-NEXT: air.channel.put async [{{.*}}]  @channel_2[%c0{{.*}}, %c1{{.*}}]
 // CHECK-NEXT: air.channel.put async [{{.*}}]  @channel_2[%c0{{.*}}, %c1{{.*}}]
+// CHECK-NEXT: air.channel.put async [{{.*}}]  @channel_2[%c1{{.*}}, %c0{{.*}}]
+// CHECK-NEXT: air.channel.put async [{{.*}}]  @channel_2[%c1{{.*}}, %c0{{.*}}]
 // CHECK-NEXT: air.channel.put async [{{.*}}]  @channel_2[%c1{{.*}}, %c1{{.*}}]
 // CHECK-NEXT: air.channel.put async [{{.*}}]  @channel_2[%c1{{.*}}, %c1{{.*}}]
 // CHECK: scf.yield

mlir/test/Transform/AIRDependencyScheduleOpt/specialize-channel-wrap-and-stride.mlir

@@ -655,8 +655,8 @@ module {
   // Scf.for on integer type.
   // CHECK-LABEL: test18
   // CHECK: air.channel.put async {{.*}} @channel_0[%c0{{.*}}, %c0{{.*}}] (%{{.*}}[%c0{{.*}}, %c0{{.*}}, %c0{{.*}}, %c0{{.*}}] [%c32{{.*}}, %c32{{.*}}, %c4{{.*}}, %c2{{.*}}] [%c512{{.*}}, %c2{{.*}}, %c128{{.*}}, %c1{{.*}}])
-  // CHECK: air.channel.put async {{.*}} @channel_0[%c1{{.*}}, %c0{{.*}}] (%{{.*}}[%c0{{.*}}, %c0{{.*}}, %c0{{.*}}, %c16384{{.*}}] [%c32{{.*}}, %c32{{.*}}, %c4{{.*}}, %c2{{.*}}] [%c512{{.*}}, %c2{{.*}}, %c128{{.*}}, %c1{{.*}}])
   // CHECK: air.channel.put async {{.*}} @channel_0[%c0{{.*}}, %c1{{.*}}] (%{{.*}}[%c0{{.*}}, %c0{{.*}}, %c0{{.*}}, %c64{{.*}}] [%c32{{.*}}, %c32{{.*}}, %c4{{.*}}, %c2{{.*}}] [%c512{{.*}}, %c2{{.*}}, %c128{{.*}}, %c1{{.*}}])
+  // CHECK: air.channel.put async {{.*}} @channel_0[%c1{{.*}}, %c0{{.*}}] (%{{.*}}[%c0{{.*}}, %c0{{.*}}, %c0{{.*}}, %c16384{{.*}}] [%c32{{.*}}, %c32{{.*}}, %c4{{.*}}, %c2{{.*}}] [%c512{{.*}}, %c2{{.*}}, %c128{{.*}}, %c1{{.*}}])
   // CHECK: air.channel.put async {{.*}} @channel_0[%c1{{.*}}, %c1{{.*}}] (%{{.*}}[%c0{{.*}}, %c0{{.*}}, %c0{{.*}}, %c16448{{.*}}] [%c32{{.*}}, %c32{{.*}}, %c4{{.*}}, %c2{{.*}}] [%c512{{.*}}, %c2{{.*}}, %c128{{.*}}, %c1{{.*}}])
 
   func.func @test18(%arg0: memref<*xf32>, %arg1: memref<*xf32>) {