【Infer Symbolic Shape BUAA No.42-44】Add ops (#66930)

Author: Guanhuachen2003

paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/multiary_infer_sym.cc

@@ -782,6 +782,51 @@ bool ConcatOpInferSymbolicShape(pir::Operation *op,
   return true;
 }
 
+bool FakeQuantizeMovingAverageAbsMaxOpInferSymbolicShape(
+    pir::Operation *op, pir::InferSymbolicShapeContext *infer_context) {
+  const symbol::ShapeOrDataDimExprs &x_shape =
+      infer_context->GetShapeOrDataForValue(op->operand_source(0));
+
+  // Validate the bit_length attribute
+  int bit_length = op->attribute<pir::Int32Attribute>("bit_length").data();
+  PADDLE_ENFORCE_EQ(bit_length >= 1 && bit_length <= 16,
+                    true,
+                    phi::errors::InvalidArgument(
+                        "'bit_length' should be between 1 and 16, but "
+                        "the received is %d",
+                        bit_length));
+
+  // Set the shape for the output tensor 'out', same as input tensor 'x'
+  infer_context->SetShapeOrDataForValue(op->result(0), x_shape);
+
+  // Create a scalar shape for the other output tensors
+  symbol::TensorShapeOrDataDimExprs scalar_shape(
+      std::vector<symbol::DimExpr>{symbol::DimExpr(1)});
+
+  // Set the shape for all scalar output tensors: 'out_scale', 'out_state',
+  // 'out_accum'
+  for (size_t i = 1; i < op->num_results(); ++i) {
+    infer_context->SetShapeOrDataForValue(op->result(i), scalar_shape);
+  }
+
+  return true;
+}
+
+bool FakeQuantizeMovingAverageAbsMax_OpInferSymbolicShape(
+    pir::Operation *op, pir::InferSymbolicShapeContext *infer_context) {
+  return FakeQuantizeMovingAverageAbsMaxOpInferSymbolicShape(op, infer_context);
+}
+
+bool FakeQuantizeDequantizeMovingAverageAbsMaxOpInferSymbolicShape(
+    pir::Operation *op, pir::InferSymbolicShapeContext *infer_context) {
+  return FakeQuantizeMovingAverageAbsMaxOpInferSymbolicShape(op, infer_context);
+}
+
+bool FakeQuantizeDequantizeMovingAverageAbsMax_OpInferSymbolicShape(
+    pir::Operation *op, pir::InferSymbolicShapeContext *infer_context) {
+  return FakeQuantizeMovingAverageAbsMaxOpInferSymbolicShape(op, infer_context);
+}
+
 bool FullWithTensorOpInferSymbolicShape(
     pir::Operation *op, pir::InferSymbolicShapeContext *infer_context) {
   pir::Value operand_source = op->operand_source(1);

paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/multiary_infer_sym.h

@@ -35,6 +35,10 @@ OP_DECLARE_INFER_SYMBOLIC_SHAPE(BilinearInterp)
 OP_DECLARE_INFER_SYMBOLIC_SHAPE(Concat)
 OP_DECLARE_INFER_SYMBOLIC_SHAPE(CrossEntropyWithSoftmax)
 OP_DECLARE_INFER_SYMBOLIC_SHAPE(CrossEntropyWithSoftmax_)
+OP_DECLARE_INFER_SYMBOLIC_SHAPE(FakeQuantizeDequantizeMovingAverageAbsMax)
+OP_DECLARE_INFER_SYMBOLIC_SHAPE(FakeQuantizeDequantizeMovingAverageAbsMax_)
+OP_DECLARE_INFER_SYMBOLIC_SHAPE(FakeQuantizeMovingAverageAbsMax)
+OP_DECLARE_INFER_SYMBOLIC_SHAPE(FakeQuantizeMovingAverageAbsMax_)
 // OP_DECLARE_INFER_SYMBOLIC_SHAPE(CoalesceTensor)
 // OP_DECLARE_INFER_SYMBOLIC_SHAPE(CoalesceTensor_)
 OP_DECLARE_INFER_SYMBOLIC_SHAPE(FullWithTensor)

paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/unary_infer_sym.cc

@@ -943,6 +943,31 @@ bool FlattenOpInferSymbolicShape(
   return true;
 }
 
+bool FakeQuantizeDequantizeAbsMaxOpInferSymbolicShape(
+    pir::Operation *op, pir::InferSymbolicShapeContext *infer_context) {
+  const symbol::ShapeOrDataDimExprs &x_shape =
+      infer_context->GetShapeOrDataForValue(op->operand_source(0));
+
+  // Validate the bit_length attribute
+  int bit_length = op->attribute<pir::Int32Attribute>("bit_length").data();
+  PADDLE_ENFORCE_EQ(bit_length >= 1 && bit_length <= 16,
+                    true,
+                    phi::errors::InvalidArgument(
+                        "'bit_length' should be between 1 and 16, but "
+                        "the received is %d",
+                        bit_length));
+
+  // Set the shape for the output tensor 'out', same as input tensor 'x'
+  infer_context->SetShapeOrDataForValue(op->result(0), x_shape);
+
+  // Set the shape for the output tensor 'out_scale' as a scalar {1}
+  symbol::TensorShapeOrDataDimExprs scalar_shape(
+      std::vector<symbol::DimExpr>{symbol::DimExpr(1)});
+  infer_context->SetShapeOrDataForValue(op->result(1), scalar_shape);
+
+  return true;
+}
+
 bool Flatten_OpInferSymbolicShape(
     pir::Operation *op, pir::InferSymbolicShapeContext *infer_context) {
   return FlattenOpInferSymbolicShape(op, infer_context);

paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/unary_infer_sym.h

@@ -62,6 +62,7 @@ OP_DECLARE_INFER_SYMBOLIC_SHAPE(FillDiagonal)
 OP_DECLARE_INFER_SYMBOLIC_SHAPE(FillDiagonal_)
 OP_DECLARE_INFER_SYMBOLIC_SHAPE(Flatten)
 OP_DECLARE_INFER_SYMBOLIC_SHAPE(Flatten_)
+OP_DECLARE_INFER_SYMBOLIC_SHAPE(FakeQuantizeDequantizeAbsMax)
 OP_DECLARE_INFER_SYMBOLIC_SHAPE(Inverse)
 // OP_DECLARE_INFER_SYMBOLIC_SHAPE(GumbelSoftmax)
 OP_DECLARE_INFER_SYMBOLIC_SHAPE(IdentityLoss)

paddle/phi/ops/yaml/ops.yaml

@@ -1715,6 +1715,7 @@
     func : fake_quantize_dequantize_abs_max
     data_type : x
   backward : fake_quantize_dequantize_abs_max_grad
+  interfaces : paddle::dialect::InferSymbolicShapeInterface
 
 - op : fake_quantize_dequantize_moving_average_abs_max
   args : (Tensor x, Tensor in_scale, Tensor in_accum, Tensor in_state, float moving_rate = 0.9, int bit_length = 8, bool is_test = false, int round_type = 1)
@@ -1727,6 +1728,7 @@
   optional : in_accum, in_state, out_state, out_accum
   backward : fake_quantize_dequantize_moving_average_abs_max_grad
   inplace: (in_scale -> out_scale)
+  interfaces : paddle::dialect::InferSymbolicShapeInterface
 
 - op : fake_quantize_moving_average_abs_max
   args : (Tensor x, Tensor in_scale, Tensor in_accum, Tensor in_state, float moving_rate = 0.9, int bit_length = 8, bool is_test = false, int round_type = 1)
@@ -1738,6 +1740,7 @@
     data_type : x
   optional : in_accum, in_state, out_state, out_accum
   inplace: (in_scale -> out_scale)
+  interfaces : paddle::dialect::InferSymbolicShapeInterface
 
 - op : fake_quantize_range_abs_max
   args : (Tensor x, Tensor in_scale, Tensor iter, int window_size = 10000,  int bit_length = 8, bool is_test = false, int round_type = 1)