[tosa] : Add e2e support for quantized matmul. (#4371)

This PR enables e2e test for quantized torch.mm and it's other variants
through the `tosa` path.

`torch` IR for quantized matmul is shown in the following snippet:

```
%2 = torch.aten._make_per_tensor_quantized_tensor %0, %float2.150000e-02, %int-25 : !torch.vtensor<[3,4],si8>, !torch.float, !torch.int -> !torch.vtensor<[3,4],!torch.qint8>
  %3 = torch.aten._make_per_tensor_quantized_tensor %1, %float1.760000e-02, %int18 : !torch.vtensor<[4,3],si8>, !torch.float, !torch.int -> !torch.vtensor<[4,3],!torch.qint8>
  %4 = torch.aten.mm %2, %3 : !torch.vtensor<[3,4],!torch.qint8>, !torch.vtensor<[4,3],!torch.qint8> -> !torch.vtensor<[3,3],!torch.qint32>
  %5 = torch.aten.int_repr %4 : !torch.vtensor<[3,3],!torch.qint32> -> !torch.vtensor<[3,3],si32>
  %6 = torch.aten._make_per_tensor_quantized_tensor %5, %float3.784000e-04, %int0 : !torch.vtensor<[3,3],si32>, !torch.float, !torch.int -> !torch.vtensor<[3,3],!torch.qint32>
  %7 = torch.aten.dequantize.tensor %6 : !torch.vtensor<[3,3],!torch.qint32> -> !torch.vtensor<[3,3],f32>
```

1. This change adds legalizations for
`_make_per_tensor_quantized_tensor`, `int_repr` which are basically cast
operations. The former op carries the zero-point/scale information for
(de)quantizing values.
2. Legalization for `dequantize.tensor` is also added which is the usual
dequantization op.
3. Legalization for `matmul` is fixed to infer the zero-point
information from the source `_make_per_tensor_quantized_tensor` ops for
the `matmul` operands. Scale doesn't need to be considered, as it will
be taken care of correctly at the output via `FuseQuantizedOps`
transform.

Author: sahas3

include/torch-mlir/Conversion/Utils/Utils.h

@@ -18,6 +18,16 @@ namespace mlir {
 namespace torch {
 namespace Torch {
 
+// Define constants
+// Float 16 limits
+constexpr float Float16Max = 65504.0f;
+constexpr float Float16Lowest = -65504.0f;
+
+// BFloat 16 limits
+constexpr float BFloat16Max = 3.38953139e38f;
+constexpr float BFloat16Lowest = -3.38953139e38f;
+
+// Define utility methods
 LogicalResult verifyLinalgCompatibleTypes(Operation *op,
                                           PatternRewriter &rewriter);
 
@@ -107,13 +117,8 @@ FailureOr<Value> unsqueezeTensor(PatternRewriter &rewriter, Operation *op,
 FailureOr<Value> squeezeTensor(PatternRewriter &rewriter, Operation *op,
                                Value input, int64_t dim);
 
-// Float 16 limits
-constexpr float Float16Max = 65504.0f;
-constexpr float Float16Lowest = -65504.0f;
+void getZeroPoint(Value value, Value &zeropoint);
 
-// BFloat 16 limits
-constexpr float BFloat16Max = 3.38953139e38f;
-constexpr float BFloat16Lowest = -3.38953139e38f;
 } // namespace Torch
 } // namespace torch
 } // namespace mlir

lib/Conversion/TorchToLinalg/Linear.cpp

@@ -28,12 +28,6 @@ using namespace mlir::torch::Torch;
 
 namespace {
 
-static void getZeroPoint(Value value, Value &zeropoint) {
-  if (auto make = value.getDefiningOp<Aten_MakePerTensorQuantizedTensorOp>()) {
-    zeropoint = make.getZeroPoint();
-  }
-}
-
 // for uint8 types, we shift down by 128 so that we can faithfully
 // represent the quantization with signed i8 types.
 static void signShift(PatternRewriter &rewriter, Location loc, Value &arg,