[torch] Fix attention on linalg for dynamic shapes (#3714)

Current version does not work for a mixture of dynamic and static shaped
batch dimensions. Rework to grab the correct dynamic shapes.

---------

Co-authored-by: dan <[email protected]>

Author: rsuderman

lib/Conversion/TorchToTMTensor/TorchToTMTensor.cpp

@@ -1607,35 +1607,23 @@ class ConvertAtenScaledDotProductAttentionOp
             op.getLoc(), "expected no attention mask when isCausal is true");
       }
 
-      SmallVector<OpFoldResult> maskSizes;
-
-      if (queryTy.hasStaticShape() && keyTy.hasStaticShape()) {
-        auto seqLenQ =
-            rewriter.getIndexAttr(queryTy.getDimSize(queryTy.getRank() - 2));
-        auto seqLenK =
-            rewriter.getIndexAttr(keyTy.getDimSize(keyTy.getRank() - 2));
-        maskSizes = {seqLenQ, seqLenK};
-        for (int i = queryTy.getRank() - 3; i >= 0; --i) {
-          auto batchSize = rewriter.getIndexAttr(queryTy.getDimSize(i));
-          maskSizes.insert(maskSizes.begin(), batchSize);
-        }
-      } else { // Dynamic shape case: <?x?x...x?xf32> for example
-        for (int i = 0; i < queryTy.getRank() - 2; ++i) {
-          Value batchSize =
-              rewriter.create<tensor::DimOp>(op.getLoc(), query, i);
-          maskSizes.push_back(batchSize);
-        }
-        Value seqLenQ = rewriter.create<tensor::DimOp>(op.getLoc(), query,
-                                                       queryTy.getRank() - 2);
-        Value seqLenK = rewriter.create<tensor::DimOp>(op.getLoc(), key,
-                                                       keyTy.getRank() - 2);
-        maskSizes.push_back(seqLenQ);
-        maskSizes.push_back(seqLenK);
+      SmallVector<int64_t> maskStatic;
+      SmallVector<Value> maskDyn;
+      for (int i = 0, s = queryTy.getRank() - 1; i < s; ++i) {
+        maskStatic.push_back(queryTy.getDimSize(i));
+        if (maskStatic.back() == ShapedType::kDynamic)
+          maskDyn.push_back(
+              rewriter.create<tensor::DimOp>(op.getLoc(), query, i));
       }
 
+      maskStatic.push_back(keyTy.getDimSize(keyTy.getRank() - 2));
+      if (maskStatic.back() == ShapedType::kDynamic)
+        maskDyn.push_back(rewriter.create<tensor::DimOp>(op.getLoc(), key,
+                                                         keyTy.getRank() - 2));
+
       Type maskType = getElementTypeOrSelf(queryTy);
-      Value emptyMask =
-          rewriter.create<tensor::EmptyOp>(op.getLoc(), maskSizes, maskType);
+      Value emptyMask = rewriter.create<tensor::EmptyOp>(
+          op.getLoc(), maskStatic, maskType, maskDyn);
 
       Value zero = rewriter.create<arith::ConstantOp>(
           op.getLoc(),

projects/pt1/e2e_testing/xfail_sets.py

@@ -37,6 +37,7 @@
         # WORKS FOR TORCH VERSION 2.5.0.dev20240902, REMOVE WHEN ENABLE_GQA IS PUT IN STABLE
         "ScaledDotProductAttentionBoolMaskModule_basic",
         "ScaledDotProductAttentionDifferentCausalModule_basic",
+        "ScaledDotProductAttentionDifferentDynamicCausalModule_basic",
         "ScaledDotProductAttentionDifferentModule_basic",
         "ScaledDotProductAttentionMaskModule_basic",
         "ScaledDotProductAttentionSameCausalModule_basic",
@@ -833,6 +834,7 @@
     "SafeSoftmaxNonNoneDtypeModule_basic",
     # REMOVE WHEN ENABLE_GQA IS ADDED
     "ScaledDotProductAttentionBoolMaskModule_basic",
+    "ScaledDotProductAttentionDifferentDynamicCausalModule_basic",
     "ScaledDotProductAttentionDifferentCausalModule_basic",
     "ScaledDotProductAttentionDifferentModule_basic",
     "ScaledDotProductAttentionMaskModule_basic",
@@ -3176,6 +3178,7 @@
     # REMOVE WHEN ENABLE_GQA IS ADDED
     "ScaledDotProductAttentionBoolMaskModule_basic",
     "ScaledDotProductAttentionDifferentCausalModule_basic",
+    "ScaledDotProductAttentionDifferentDynamicCausalModule_basic",
     "ScaledDotProductAttentionSameCausalModule_basic",
     "ScatterAddStaticModule_basic",
     "TensorsConcatComplex128FloatModule_basic",
@@ -4679,6 +4682,7 @@
     "ScalarImplicitIntModule_basic",
     # REMOVE WHEN ENABLE_GQA IS ADDED
     "ScaledDotProductAttentionBoolMaskModule_basic",
+    "ScaledDotProductAttentionDifferentDynamicCausalModule_basic",
     "ScaledDotProductAttentionSameCausalModule_basic",
     "ScaledDotProductAttentionSameDynamicModule_basic",
     "ScatterReduceFloatMaxModule",

projects/pt1/python/torch_mlir_e2e_test/test_suite/basic.py

@@ -5370,6 +5370,35 @@ def forward(self, query, key, value):
 @register_test_case(
     module_factory=lambda: ScaledDotProductAttentionDifferentCausalModule()
 )
+def ScaledDotProductAttentionDifferentDynamicCausalModule_basic(module, tu: TestUtils):
+    query = torch.randn(2, 3, 8, 16, dtype=torch.float32)
+    key = torch.randn(2, 3, 12, 16, dtype=torch.float32)
+    value = torch.randn(2, 3, 12, 20, dtype=torch.float32)
+    module.forward(query, key, value)
+
+
+class ScaledDotProductAttentionDifferentDynamicCausalModule(torch.nn.Module):
+    def __init__(self):
+        super().__init__()
+
+    @export
+    @annotate_args(
+        [
+            None,
+            ([2, 3, -1, 16], torch.float32, True),
+            ([2, 3, -1, 16], torch.float32, True),
+            ([2, 3, -1, 20], torch.float32, True),
+        ]
+    )
+    def forward(self, query, key, value):
+        return torch.ops.aten.scaled_dot_product_attention(
+            query, key, value, is_causal=True
+        )
+
+
+@register_test_case(
+    module_factory=lambda: ScaledDotProductAttentionDifferentDynamicCausalModule()
+)
 def ScaledDotProductAttentionDifferentCausalModule_basic(module, tu: TestUtils):
     query = torch.randn(2, 3, 8, 16, dtype=torch.float32)
     key = torch.randn(2, 3, 12, 16, dtype=torch.float32)