[lib/conversion] Create seed only if needed in `convert-torch-convers… (

#3926)

…ion-to-mlprogram` pass

This PR changes `convert-torch-conversion-to-mlprogram` pass
implementation by moving seed generation inside `ConvertGetNextSeedOp`
pattern.
Previously, global seed was being created by this pass, even when its
only consumer `torch_c.get_next_seed` op is not present in the IR. This
pass is part of Torch->Linalg conversion pipeline. Always creating
global seed created an issue for the case when downstream compiler
doesn't expect/support `ml_program` dialect in linalg on tensor IR
format. However, when starting torch IR has `torch_c.get_next_seed` op,
`ml_program` will still be present and will need to be handled by
downstream compilers.

lib/Conversion/TorchConversionToMLProgram/TorchConversionToMLProgram.cpp

@@ -59,6 +59,13 @@ class ConvertGetNextSeedOp : public OpConversionPattern<GetNextSeedOp> {
   matchAndRewrite(GetNextSeedOp op, OpAdaptor adaptor,
                   ConversionPatternRewriter &rewriter) const override {
     Location loc = op.getLoc();
+
+    // Check for global seed and create if it doesn't exist.
+    auto module = op->getParentOfType<ModuleOp>();
+    OpBuilder b(module.getBodyRegion());
+    if (failed(getOrCreateGlobalVariableForSeed(b, module)))
+      return failure();
+
     // Generate sequence for getting the next seed with LCG step:
     //    nextSeed = (multiplier * currentSeed + incrementStep) mod 2^64.
     // Refer to https://en.wikipedia.org/wiki/Linear_congruential_generator.
@@ -115,11 +122,6 @@ class ConvertTorchConversionToMLProgram
     typeConverter.addConversion([](Type type) { return type; });
     TorchConversion::setupBackendTypeConversion(target, typeConverter);
 
-    auto module = getOperation();
-    OpBuilder b(module.getBodyRegion());
-    if (failed(getOrCreateGlobalVariableForSeed(b, module)))
-      signalPassFailure();
-
     RewritePatternSet patterns(context);
     target.addIllegalOp<GetNextSeedOp>();
     patterns.add<ConvertGetNextSeedOp>(typeConverter, context);

test/Conversion/TorchConversionToMLProgram/basic.mlir

@@ -17,3 +17,16 @@ module {
     return %seed : i64
   }
 }
+
+// -----
+
+module {
+  func.func @no_seed_needed(%arg0: tensor<2x3xf32>) -> !torch.vtensor<[2,3],f32> {
+    %0 = torch_c.from_builtin_tensor %arg0 : tensor<2x3xf32> -> !torch.vtensor<[2,3],f32>
+    return %0 : !torch.vtensor<[2,3],f32>
+  }
+}
+
+// CHECK-NOT: ml_program.global
+// CHECK-LABEL: @no_seed_needed
+// CHECK-NEXT: torch_c.from_builtin_tensor

test/Conversion/TorchConversionToMLProgram/multiple_functions.mlir

@@ -11,5 +11,5 @@ module {
   func.func private @f7() -> i64
 }
 
-// CHECK:     ml_program.global private mutable @global_seed(dense<0> : tensor<i64>) : tensor<i64>
+// CHECK-NOT:     ml_program.global private mutable @global_seed(dense<0> : tensor<i64>) : tensor<i64>
 // CHECK-NOT: @global_seed