Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[ONNX] Add support for non-scalar zero points for Onnx.MatmulInteger op #4036

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[ONNX] Add support for non-scalar zero points for Onnx.MatmulInteger op #4036

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
88 changes: 72 additions & 16 deletions lib/Conversion/TorchOnnxToTorch/DefaultDomainGtoP.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -568,36 +568,92 @@ void mlir::torch::onnx_c::populateDefaultDomainGtoP(
rewriter.getIntegerAttr(rewriter.getIntegerType(64), 0));
}

bool isChannelQuantizationForLhs = false;
if (auto zpTy = dyn_cast<Torch::ValueTensorType>(lhsZp.getType())) {
for (auto dim : zpTy.getSizes())
if (dim != 1)
auto lhsZpSize = zpTy.getSizes();
if (lhsZpSize.size() == 0 ||
llvm::all_of(lhsZpSize, [](int64_t d) { return d == 1; })) {
lhsZp = rewriter.create<Torch::AtenItemOp>(
binder.getLoc(), rewriter.getType<Torch::IntType>(), lhsZp);
} else if (lhsZpSize.size() == 1) {
auto lhsSize = lhsTy.getSizes();
if (lhsSize.size() != 2 || lhsSize[0] != lhsZpSize[0])
return failure();
lhsZp = rewriter.create<Torch::AtenItemOp>(
binder.getLoc(), rewriter.getType<Torch::IntType>(), lhsZp);
isChannelQuantizationForLhs = true;
} else {
return failure();
}
}

bool isChannelQuantizationForRhs = false;
if (auto zpTy = dyn_cast<Torch::ValueTensorType>(rhsZp.getType())) {
for (auto dim : zpTy.getSizes())
if (dim != 1)
auto rhsZpSize = zpTy.getSizes();
if (rhsZpSize.size() == 0 ||
llvm::all_of(rhsZpSize, [](int64_t d) { return d == 1; })) {
rhsZp = rewriter.create<Torch::AtenItemOp>(
binder.getLoc(), rewriter.getType<Torch::IntType>(), rhsZp);
} else if (rhsZpSize.size() == 1) {
auto rhsSize = rhsTy.getSizes();
if (rhsSize.size() != 2 || rhsSize[1] != rhsZpSize[0])
return failure();
rhsZp = rewriter.create<Torch::AtenItemOp>(
binder.getLoc(), rewriter.getType<Torch::IntType>(), rhsZp);
isChannelQuantizationForRhs = true;
} else {
return failure();
}
}

Value scale = rewriter.create<Torch::ConstantFloatOp>(
binder.getLoc(), rewriter.getType<Torch::FloatType>(),
rewriter.getF64FloatAttr(1.0));

auto lhsQTy = getQTorchTypeFromTorchIntType(lhsTy);
auto rhsQTy = getQTorchTypeFromTorchIntType(rhsTy);

if (!lhsQTy || !rhsQTy)
return rewriter.notifyMatchFailure(binder.op, "failed to get qtype");

lhs = rewriter.create<Torch::Aten_MakePerTensorQuantizedTensorOp>(
binder.getLoc(), lhsQTy, lhs, scale, lhsZp);
rhs = rewriter.create<Torch::Aten_MakePerTensorQuantizedTensorOp>(
binder.getLoc(), rhsQTy, rhs, scale, rhsZp);
Value f32Ty = rewriter.create<Torch::ConstantIntOp>(
binder.getLoc(), rewriter.getI64IntegerAttr(
(int64_t)torch_upstream::ScalarType::Float));
Value none = rewriter.create<Torch::ConstantNoneOp>(binder.getLoc());

if (isChannelQuantizationForLhs) {
Value axis = rewriter.create<Torch::ConstantIntOp>(
binder.getLoc(), rewriter.getType<Torch::IntType>(),
rewriter.getI64IntegerAttr(0));
Torch::ValueTensorType lhsZpTy =
dyn_cast<Torch::ValueTensorType>(lhsZp.getType());
Type scaleTy = lhsZpTy.getWithSizesAndDtype(lhsZpTy.getSizes(),
rewriter.getF32Type());
Value scale = rewriter.create<Torch::AtenOnesLikeOp>(
binder.getLoc(), scaleTy, /*self=*/lhsZp, f32Ty, /*layout=*/none,
/*device=*/none, /*pin_memory=*/none, /*memory_format=*/none);
lhs = rewriter.create<Torch::Aten_MakePerChannelQuantizedTensorOp>(
binder.getLoc(), lhsQTy, lhs, scale, lhsZp, axis);
} else {
Value scale = rewriter.create<Torch::ConstantFloatOp>(
binder.getLoc(), rewriter.getType<Torch::FloatType>(),
rewriter.getF64FloatAttr(1.0));
lhs = rewriter.create<Torch::Aten_MakePerTensorQuantizedTensorOp>(
binder.getLoc(), lhsQTy, lhs, scale, lhsZp);
}

if (isChannelQuantizationForRhs) {
Value axis = rewriter.create<Torch::ConstantIntOp>(
binder.getLoc(), rewriter.getType<Torch::IntType>(),
rewriter.getI64IntegerAttr(1));
Torch::ValueTensorType rhsZpTy =
dyn_cast<Torch::ValueTensorType>(rhsZp.getType());
Type scaleTy = rhsZpTy.getWithSizesAndDtype(rhsZpTy.getSizes(),
rewriter.getF32Type());
Value scale = rewriter.create<Torch::AtenOnesLikeOp>(
binder.getLoc(), scaleTy, /*self=*/rhsZp, f32Ty, /*layout=*/none,
/*device=*/none, /*pin_memory=*/none, /*memory_format=*/none);
rhs = rewriter.create<Torch::Aten_MakePerChannelQuantizedTensorOp>(
binder.getLoc(), rhsQTy, rhs, scale, rhsZp, axis);
} else {
Value scale = rewriter.create<Torch::ConstantFloatOp>(
binder.getLoc(), rewriter.getType<Torch::FloatType>(),
rewriter.getF64FloatAttr(1.0));
rhs = rewriter.create<Torch::Aten_MakePerTensorQuantizedTensorOp>(
binder.getLoc(), rhsQTy, rhs, scale, rhsZp);
}

rewriter.replaceOpWithNewOp<Torch::AtenMatmulOp>(binder.op, resultType,
lhs, rhs);
Expand Down
59 changes: 53 additions & 6 deletions test/Conversion/TorchOnnxToTorch/simple_ops_g_to_p.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -569,9 +569,10 @@ func.func @test_matmulinteger(%arg0: !torch.vtensor<[4,3],ui8>, %arg1: !torch.vt
%0 = torch.operator "onnx.MatMulInteger"(%arg0, %arg1, %arg2, %arg3) : (!torch.vtensor<[4,3],ui8>, !torch.vtensor<[3,2],ui8>, !torch.vtensor<[1],ui8>, !torch.vtensor<[1],ui8>) -> !torch.vtensor<[4,2],si32>
// CHECK: %[[LITEM:.+]] = torch.aten.item %arg2
// CHECK: %[[RITEM:.+]] = torch.aten.item %arg3
// CHECK: %[[SCALE:.+]] = torch.constant.float 1.000000e+00
// CHECK: %[[LMAKE:.+]] = torch.aten._make_per_tensor_quantized_tensor %arg0, %[[SCALE]], %[[LITEM]] : !torch.vtensor<[4,3],ui8>, !torch.float, !torch.int -> !torch.vtensor<[4,3],!torch.quint8>
// CHECK: %[[RMAKE:.+]] = torch.aten._make_per_tensor_quantized_tensor %arg1, %[[SCALE]], %[[RITEM]] : !torch.vtensor<[3,2],ui8>, !torch.float, !torch.int -> !torch.vtensor<[3,2],!torch.quint8>
// CHECK: %[[L_SCALE:.+]] = torch.constant.float 1.000000e+00
// CHECK: %[[LMAKE:.+]] = torch.aten._make_per_tensor_quantized_tensor %arg0, %[[L_SCALE]], %[[LITEM]] : !torch.vtensor<[4,3],ui8>, !torch.float, !torch.int -> !torch.vtensor<[4,3],!torch.quint8>
// CHECK: %[[R_SCALE:.+]] = torch.constant.float 1.000000e+00
// CHECK: %[[RMAKE:.+]] = torch.aten._make_per_tensor_quantized_tensor %arg1, %[[R_SCALE]], %[[RITEM]] : !torch.vtensor<[3,2],ui8>, !torch.float, !torch.int -> !torch.vtensor<[3,2],!torch.quint8>
// CHECK: %[[MM:.+]] = torch.aten.matmul %[[LMAKE]], %[[RMAKE]]
// CHECK: return %[[MM]]
return %0 : !torch.vtensor<[4,2],si32>
Expand All @@ -584,13 +585,59 @@ func.func @test_matmulinteger_batched(%arg0: !torch.vtensor<[7,4,3],ui8>, %arg1:
%0 = torch.operator "onnx.MatMulInteger"(%arg0, %arg1, %arg2, %arg3) : (!torch.vtensor<[7,4,3],ui8>, !torch.vtensor<[3,2],ui8>, !torch.vtensor<[1],ui8>, !torch.vtensor<[1],ui8>) -> !torch.vtensor<[7,4,2],si32>
// CHECK: %[[LITEM:.+]] = torch.aten.item %arg2
// CHECK: %[[RITEM:.+]] = torch.aten.item %arg3
// CHECK: %[[SCALE:.+]] = torch.constant.float 1.000000e+00
// CHECK: %[[LMAKE:.+]] = torch.aten._make_per_tensor_quantized_tensor %arg0, %[[SCALE]], %[[LITEM]] : !torch.vtensor<[7,4,3],ui8>, !torch.float, !torch.int -> !torch.vtensor<[7,4,3],!torch.quint8>
// CHECK: %[[RMAKE:.+]] = torch.aten._make_per_tensor_quantized_tensor %arg1, %[[SCALE]], %[[RITEM]] : !torch.vtensor<[3,2],ui8>, !torch.float, !torch.int -> !torch.vtensor<[3,2],!torch.quint8>
// CHECK: %[[L_SCALE:.+]] = torch.constant.float 1.000000e+00
// CHECK: %[[LMAKE:.+]] = torch.aten._make_per_tensor_quantized_tensor %arg0, %[[L_SCALE]], %[[LITEM]] : !torch.vtensor<[7,4,3],ui8>, !torch.float, !torch.int -> !torch.vtensor<[7,4,3],!torch.quint8>
// CHECK: %[[R_SCALE:.+]] = torch.constant.float 1.000000e+00
// CHECK: %[[RMAKE:.+]] = torch.aten._make_per_tensor_quantized_tensor %arg1, %[[R_SCALE]], %[[RITEM]] : !torch.vtensor<[3,2],ui8>, !torch.float, !torch.int -> !torch.vtensor<[3,2],!torch.quint8>
// CHECK: %[[MM:.+]] = torch.aten.matmul %[[LMAKE]], %[[RMAKE]]
// CHECK: return %[[MM]]
return %0 : !torch.vtensor<[7,4,2],si32>
}

// -----

// CHECK-LABEL: func.func @test_matmulinteger_non_scalar_lhsZp(
// CHECK-SAME: %[[VAL_0:.*]]: !torch.vtensor<[16,2],ui8>,
// CHECK-SAME: %[[VAL_1:.*]]: !torch.vtensor<[2,768],si8>,
// CHECK-SAME: %[[VAL_2:.*]]: !torch.vtensor<[16],ui8>,
// CHECK-SAME: %[[VAL_3:.*]]: !torch.vtensor<[],si8>) -> !torch.vtensor<[16,768],si32> attributes {torch.onnx_meta.ir_version = 7 : si64, torch.onnx_meta.opset_version = 21 : si64, torch.onnx_meta.producer_name = "pytorch", torch.onnx_meta.producer_version = "0.1.0"} {
func.func @test_matmulinteger_non_scalar_lhsZp(%arg0: !torch.vtensor<[16, 2],ui8>, %arg1: !torch.vtensor<[2,768],si8>, %arg2: !torch.vtensor<[16],ui8>, %arg3: !torch.vtensor<[],si8>) -> !torch.vtensor<[16,768],si32> attributes {torch.onnx_meta.ir_version = 7 : si64, torch.onnx_meta.opset_version = 21 : si64, torch.onnx_meta.producer_name = "pytorch", torch.onnx_meta.producer_version = "0.1.0"} {
// CHECK: %[[VAL_4:.*]] = torch.aten.item %[[VAL_3]] : !torch.vtensor<[],si8> -> !torch.int
// CHECK: %[[VAL_5:.*]] = torch.constant.int 6
// CHECK: %[[VAL_6:.*]] = torch.constant.none
// CHECK: %[[VAL_7:.*]] = torch.constant.int 0
// CHECK: %[[VAL_8:.*]] = torch.aten.ones_like %[[VAL_2]], %[[VAL_5]], %[[VAL_6]], %[[VAL_6]], %[[VAL_6]], %[[VAL_6]] : !torch.vtensor<[16],ui8>, !torch.int, !torch.none, !torch.none, !torch.none, !torch.none -> !torch.vtensor<[16],f32>
// CHECK: %[[VAL_9:.*]] = torch.aten._make_per_channel_quantized_tensor %[[VAL_0]], %[[VAL_8]], %[[VAL_2]], %[[VAL_7]] : !torch.vtensor<[16,2],ui8>, !torch.vtensor<[16],f32>, !torch.vtensor<[16],ui8>, !torch.int -> !torch.vtensor<[16,2],!torch.quint8>
// CHECK: %[[VAL_10:.*]] = torch.constant.float 1.000000e+00
// CHECK: %[[VAL_11:.*]] = torch.aten._make_per_tensor_quantized_tensor %[[VAL_1]], %[[VAL_10]], %[[VAL_4]] : !torch.vtensor<[2,768],si8>, !torch.float, !torch.int -> !torch.vtensor<[2,768],!torch.qint8>
// CHECK: %[[VAL_12:.*]] = torch.aten.matmul %[[VAL_9]], %[[VAL_11]] : !torch.vtensor<[16,2],!torch.quint8>, !torch.vtensor<[2,768],!torch.qint8> -> !torch.vtensor<[16,768],si32>
// CHECK: return %[[VAL_12]] : !torch.vtensor<[16,768],si32>
%0 = torch.operator "onnx.MatMulInteger"(%arg0, %arg1, %arg2, %arg3) : (!torch.vtensor<[16,2],ui8>, !torch.vtensor<[2,768],si8>, !torch.vtensor<[16],ui8>, !torch.vtensor<[],si8>) -> !torch.vtensor<[16,768],si32>
return %0 : !torch.vtensor<[16,768],si32>
}

// -----

// CHECK-LABEL: func.func @test_matmulinteger_non_scalar_rhsZp(
// CHECK-SAME: %[[VAL_0:.*]]: !torch.vtensor<[?,?],ui8>,
// CHECK-SAME: %[[VAL_1:.*]]: !torch.vtensor<[2,768],si8>,
// CHECK-SAME: %[[VAL_2:.*]]: !torch.vtensor<[],ui8>,
// CHECK-SAME: %[[VAL_3:.*]]: !torch.vtensor<[768],si8>) -> !torch.vtensor<[?,768],si32> attributes {torch.onnx_meta.ir_version = 7 : si64, torch.onnx_meta.opset_version = 21 : si64, torch.onnx_met
func.func @test_matmulinteger_non_scalar_rhsZp(%arg0: !torch.vtensor<[?,?],ui8>, %arg1: !torch.vtensor<[2,768],si8>, %arg2: !torch.vtensor<[],ui8>, %arg3: !torch.vtensor<[768],si8>) -> !torch.vtensor<[?,768],si32> attributes {torch.onnx_meta.ir_version = 7 : si64, torch.onnx_meta.opset_version = 21 : si64, torch.onnx_meta.producer_name = "pytorch", torch.onnx_meta.producer_version = "0.1.0"} {
// CHECK: %[[VAL_4:.*]] = torch.aten.item %[[VAL_2]] : !torch.vtensor<[],ui8> -> !torch.int
// CHECK: %[[VAL_5:.*]] = torch.constant.int 6
// CHECK: %[[VAL_6:.*]] = torch.constant.none
// CHECK: %[[VAL_7:.*]] = torch.constant.float 1.000000e+00
// CHECK: %[[VAL_8:.*]] = torch.aten._make_per_tensor_quantized_tensor %[[VAL_0]], %[[VAL_7]], %[[VAL_4]] : !torch.vtensor<[?,?],ui8>, !torch.float, !torch.int -> !torch.vtensor<[?,?],!torch.quint8>
// CHECK: %[[VAL_9:.*]] = torch.constant.int 1
// CHECK: %[[VAL_10:.*]] = torch.aten.ones_like %[[VAL_3]], %[[VAL_5]], %[[VAL_6]], %[[VAL_6]], %[[VAL_6]], %[[VAL_6]] : !torch.vtensor<[768],si8>, !torch.int, !torch.none, !torch.none, !torch.none, !torch.none -> !torch.vtensor<[768],f32>
// CHECK: %[[VAL_11:.*]] = torch.aten._make_per_channel_quantized_tensor %[[VAL_1]], %[[VAL_10]], %[[VAL_3]], %[[VAL_9]] : !torch.vtensor<[2,768],si8>, !torch.vtensor<[768],f32>, !torch.vtensor<[768],si8>, !torch.int -> !torch.vtensor<[2,768],!torch.qint8>
// CHECK: %[[VAL_12:.*]] = torch.aten.matmul %[[VAL_8]], %[[VAL_11]] : !torch.vtensor<[?,?],!torch.quint8>, !torch.vtensor<[2,768],!torch.qint8> -> !torch.vtensor<[?,768],si32>
// CHECK: return %[[VAL_12]] : !torch.vtensor<[?,768],si32>
%0 = torch.operator "onnx.MatMulInteger"(%arg0, %arg1, %arg2, %arg3) : (!torch.vtensor<[?,?],ui8>, !torch.vtensor<[2,768],si8>, !torch.vtensor<[],ui8>, !torch.vtensor<[768],si8>) -> !torch.vtensor<[?,768],si32>
return %0 : !torch.vtensor<[?,768],si32>
}

// -----

// CHECK-LABEL: func.func @test_mul
Expand Down
Loading