fix(imperative/astype): fix astype when target dtype is int4

GitOrigin-RevId: 1c5195176df3afdbf5ff1864d5e696b849fcc6cb

imperative/python/test/unit/core/test_dtype_quant.py

@@ -85,54 +85,6 @@ def _check_result_attr(oup, dtype, dtype_str, is_unsigned=True):
         np.testing.assert_equal(get_zero_point(oup.dtype), get_zero_point(dtype))
 
 
-def test_dtype_int8_ffi_handle():
-    device = "xpux"
-    shape = (3, 3, 3)
-    data = np.random.random(shape).astype(np.float32) * 5 - 1
-
-    def identity(x):
-        return x
-
-    dtype = quint8(0.01, 127)
-    inp = convert_to_quint8(data, dtype)
-    oup = _get_compiled_result(inp, dtype, shape, device, calc_func=identity)
-    _check_result_attr(oup, dtype, "quint8")
-    np.testing.assert_allclose(convert_from_quint8(oup), convert_from_quint8(inp))
-
-    dtype = qint8(0.01)
-    inp = convert_to_qint8(data, dtype)
-    oup = _get_compiled_result(inp, dtype, shape, device, calc_func=identity)
-    _check_result_attr(oup, dtype, "qint8", is_unsigned=False)
-    np.testing.assert_allclose(convert_from_qint8(oup), convert_from_qint8(inp))
-
-
-def test_quint8_typecvt():
-    device = "xpux"
-    shape = (3, 3, 3)
-    data = np.random.random(shape).astype(np.float32) * 5 - 1
-
-    def typecvt(x, dt=None):
-        (y,) = G.apply_normal_varnode(ops.TypeCvt(dtype=dt), x)
-        return y
-
-    # convert to quint8
-    dtype = quint8(0.01, 135)
-    oup = _get_compiled_result(
-        data, np.float32, shape, device, calc_func=partial(typecvt, dt=dtype)
-    )
-    _check_result_attr(oup, dtype, "quint8")
-    np.testing.assert_equal(oup, convert_to_quint8(data, dtype))
-
-    # convert from quint8 to float32
-    oup_float = _get_compiled_result(
-        oup, dtype, shape, device, calc_func=partial(typecvt, dt=np.float32)
-    )
-    assert oup_float.dtype == np.float32
-    np.testing.assert_equal(
-        oup_float, convert_from_quint8(convert_to_quint8(data, dtype))
-    )
-
-
 def test_dtype_quint4():
     with pytest.raises(ValueError):
         blah = quint4(0.05, 0.233)
@@ -161,31 +113,43 @@ def test_dtype_qint4():
     np.testing.assert_allclose(get_scale(dt), 0.01)
 
 
-def test_dtype_int4_ffi_handle():
-    device = "xpux"
-    shape = (3, 3, 3)
-    data = np.random.random(shape).astype(np.float32) * 5 - 1
-
+@pytest.mark.parametrize(
+    "dtype, dtype_name",
+    [
+        (quint4(0.01, 5), "quint4"),
+        (qint4(0.01), "qint4"),
+        (quint8(0.01, 135), "quint8"),
+        (qint8(0.01), "qint8"),
+    ],
+)
+def test_dtype_qint_mgb_ffi_handle(dtype, dtype_name):
     def identity(x):
         return x
 
-    dtype = quint4(0.01, 7)
-    inp = convert_to_quint4(data, dtype)
-    oup = _get_compiled_result(inp, dtype, shape, device, calc_func=identity)
-    _check_result_attr(oup, dtype, "quint4")
-    np.testing.assert_allclose(convert_from_quint4(oup), convert_from_quint4(inp))
+    convert_to_dtype = eval("convert_to_%s" % dtype_name)
+    convert_from_dtype = eval("convert_from_%s" % dtype_name)
+    device = "xpux"
+    shape = (3, 3, 3)
+    data = np.random.random(shape).astype(np.float32) * 5 - 1
 
-    dtype = qint4(0.01)
-    inp = convert_to_qint4(data, dtype)
+    inp = convert_to_dtype(data, dtype)
     oup = _get_compiled_result(inp, dtype, shape, device, calc_func=identity)
-    _check_result_attr(oup, dtype, "qint4", is_unsigned=False)
-    np.testing.assert_allclose(convert_from_qint4(oup), convert_from_qint4(inp))
-
-
-@pytest.mark.skipif(
-    get_device_count("gpu") != 0, reason="TypeCvt to quint4 is not supported on GPU",
+    _check_result_attr(oup, dtype, dtype_name, dtype_name.startswith("qu"))
+    np.testing.assert_allclose(convert_from_dtype(oup), convert_from_dtype(inp))
+
+
+@pytest.mark.parametrize(
+    "dtype, dtype_name",
+    [
+        (quint4(0.01, 5), "quint4"),
+        (qint4(0.01), "qint4"),
+        (quint8(0.01, 135), "quint8"),
+        (qint8(0.01), "qint8"),
+    ],
 )
-def test_quint4_typecvt():
+def test_qint_typecvt(dtype, dtype_name):
+    convert_to_dtype = eval("convert_to_%s" % dtype_name)
+    convert_from_dtype = eval("convert_from_%s" % dtype_name)
     device = "xpux"
     shape = (3, 3, 3)
     data = np.random.random(shape).astype(np.float32) * 5 - 1
@@ -195,12 +159,11 @@ def test_quint4_typecvt():
         return y
 
     # convert to quint4
-    dtype = quint4(0.01, 5)
     oup = _get_compiled_result(
         data, np.float32, shape, device, calc_func=partial(typecvt, dt=dtype)
     )
-    _check_result_attr(oup, dtype, "quint4")
-    np.testing.assert_equal(oup, convert_to_quint4(data, dtype))
+    _check_result_attr(oup, dtype, dtype_name, dtype_name.startswith("qu"))
+    np.testing.assert_equal(oup, convert_to_dtype(data, dtype))
 
     # convert from quint4 to float32
     oup_float = _get_compiled_result(
@@ -208,5 +171,62 @@ def test_quint4_typecvt():
     )
     assert oup_float.dtype == np.float32
     np.testing.assert_equal(
-        oup_float, convert_from_quint4(convert_to_quint4(data, dtype))
+        oup_float, convert_from_dtype(convert_to_dtype(data, dtype))
+    )
+
+
+@pytest.mark.parametrize(
+    "dtype, dtype_name",
+    [
+        (quint4(0.01, 5), "quint4"),
+        (qint4(0.01), "qint4"),
+        (quint8(0.01, 135), "quint8"),
+        (qint8(0.01), "qint8"),
+    ],
+)
+def test_qint_astype(dtype, dtype_name):
+    convert_to_dtype = eval("convert_to_%s" % dtype_name)
+    convert_from_dtype = eval("convert_from_%s" % dtype_name)
+    shape = (3, 3, 3)
+    data = np.random.random(shape).astype(np.float32) * 5 - 1
+
+    inp = Tensor(data, dtype="float32")
+    # convert to quint4
+    oup = inp.astype(dtype)
+    _check_result_attr(oup, dtype, dtype_name, dtype_name.startswith("qu"))
+    np.testing.assert_equal(oup.numpy(), convert_to_dtype(data, dtype))
+
+    # convert from quint4 to float32
+    oup_float = oup.astype("float32")
+    assert oup_float.dtype == np.float32
+    np.testing.assert_equal(
+        oup_float.numpy(), convert_from_dtype(convert_to_dtype(data, dtype))
+    )
+
+
+@pytest.mark.parametrize(
+    "dtype, dtype_name",
+    [
+        (quint4(0.01, 5), "quint4"),
+        (qint4(0.01), "qint4"),
+        (quint8(0.01, 135), "quint8"),
+        (qint8(0.01), "qint8"),
+    ],
+)
+def test_qint_new_tensor(dtype, dtype_name):
+
+    convert_to_dtype = eval("convert_to_%s" % dtype_name)
+    convert_from_dtype = eval("convert_from_%s" % dtype_name)
+    shape = (3, 3, 3)
+    data = np.random.random(shape).astype(np.float32) * 5 - 1
+    # create a new Tensor with quint8 dtype
+    inp = Tensor(convert_to_dtype(data, dtype), dtype=dtype)
+    _check_result_attr(inp, dtype, dtype_name, dtype_name.startswith("qu"))
+    np.testing.assert_equal(inp.numpy(), convert_to_dtype(data, dtype))
+
+    # convert from quint8 to float32
+    inp_float = inp.astype("float32")
+    assert inp_float.dtype == np.float32
+    np.testing.assert_equal(
+        inp_float.numpy(), convert_from_dtype(convert_to_dtype(data, dtype))
     )

imperative/python/test/unit/quantization/test_op.py

@@ -12,7 +12,7 @@ import pytest
 import megengine as mge
 import megengine.functional as F
 from megengine.core.tensor import dtype
-from megengine.device import get_device_count
+from megengine.device import get_cuda_compute_capability, get_device_count
 from megengine.functional.elemwise import _elemwise_multi_type, _elwise
 from megengine.module.quantized.conv import ConvTranspose2d
 from megengine.quantization import QuantMode, create_qparams
@@ -171,8 +171,92 @@ def test_conv_bias():
     run(10, 36, 8, 46, 26, 2, 2, 2, 1, 1, 2, True, "relu")
 
 
+@pytest.mark.skip(reason="does not support int4 when cuda version is lower than 10.2")
+def test_conv_bias_int4():
+    inp_scale = 1.5
+    w_scale = 2.5
+    outp_scale = 1.5
+    inp_dtype = dtype.quint4(inp_scale, 0)
+    w_dtype = dtype.qint4(w_scale)
+    b_dtype = dtype.qint32(inp_scale * w_scale)
+    out_dtype = dtype.quint4(outp_scale, 0)
+
+    def run(
+        N,
+        IC,
+        OC,
+        IH,
+        IW,
+        KH,
+        KW,
+        PH,
+        PW,
+        SH,
+        SW,
+        has_bias=True,
+        nonlinear_mode="identity",
+    ):
+        inp_v = np.random.normal(size=(N, IC, IH, IW))
+        w_v = np.random.normal(size=(OC, IC, KH, KW))
+        b_v = np.random.normal(size=(1, OC, 1, 1))
+        inp_scale = dtype.get_scale(inp_dtype)
+        w_scale = dtype.get_scale(w_dtype)
+        b_scale = dtype.get_scale(b_dtype)
+
+        inpv = dtype.convert_to_quint4(inp_v * inp_scale, inp_dtype)
+        wv = dtype.convert_to_qint4(w_v * w_scale, w_dtype)
+        bv = dtype.convert_to_qint32(b_v * b_scale, b_dtype)
+
+        inp_uint4 = mge.Tensor(inpv, dtype=inp_dtype)
+        w_int4 = mge.Parameter(wv, dtype=w_dtype)
+        b_int32 = mge.Parameter(bv, dtype=b_dtype)
+
+        inp_fp32 = inp_uint4.astype("float32")
+        w_fp32 = w_int4.astype("float32")
+        b_fp32 = b_int32.astype("float32")
+
+        def run_conv2d(inp, w, b):
+            O = F.conv2d(
+                inp, w, b if has_bias else None, stride=(SH, SW), padding=(PH, PW),
+            )
+            if nonlinear_mode == "relu":
+                return F.relu(O)
+            else:
+                return O
+
+        def run_conv_bias(inp, w, b):
+            b = b if has_bias else mge.Parameter(np.zeros_like(b.numpy()))
+            return F.quantized.conv_bias_activation(
+                inp,
+                w,
+                b,
+                stride=(SH, SW),
+                padding=(PH, PW),
+                dtype=out_dtype,
+                nonlinear_mode=nonlinear_mode,
+            )
+
+        expected = run_conv2d(inp_fp32, w_fp32, b_fp32)
+        expected = expected.astype(out_dtype).astype("float32")
+        result = run_conv_bias(inp_uint4, w_int4, b_int32).astype("float32")
+        expected = F.flatten(expected)
+        result = F.flatten(result)
+        np.testing.assert_allclose(result.numpy(), expected.numpy(), atol=outp_scale)
+
+    run(1, 4, 4, 24, 33, 1, 1, 2, 3, 1, 1, False)
+    run(10, 12, 24, 46, 46, 1, 1, 2, 1, 3, 1, False)
+    run(10, 36, 8, 46, 26, 2, 2, 2, 1, 1, 2, False)
+
+    run(1, 4, 4, 24, 33, 1, 1, 2, 3, 1, 1)
+    run(10, 12, 24, 46, 46, 1, 1, 2, 1, 3, 1)
+    run(10, 36, 8, 46, 26, 2, 2, 2, 1, 1, 2)
+
+    run(10, 36, 8, 46, 26, 2, 2, 2, 1, 1, 2, False, "relu")
+    run(10, 36, 8, 46, 26, 2, 2, 2, 1, 1, 2, True, "relu")
+
+
 @pytest.mark.skipif(
-    get_device_count("gpu") > 0,
+    get_cuda_compute_capability(0) < 61,
     reason="does not support int8 when gpu compute capability less than 6.1",
 )
 def test_conv_transpose2d():

imperative/python/test/unit/utils/test_network_node.py

@@ -13,7 +13,11 @@ import megengine.random as rand
 from megengine.core._imperative_rt.core2 import apply
 from megengine.core._wrap import Device
 from megengine.core.ops import builtin
-from megengine.device import get_device_count, is_cuda_available
+from megengine.device import (
+    get_cuda_compute_capability,
+    get_device_count,
+    is_cuda_available,
+)
 from megengine.functional.debug_param import (
     get_execution_strategy,
     set_execution_strategy,
@@ -287,7 +291,7 @@ def test_deformable_ps_roi_pooling():
 
 
 @pytest.mark.skipif(
-    get_device_count("gpu") > 0,
+    get_cuda_compute_capability(0) < 61,
     reason="does not support int8 when gpu compute capability less than 6.1",
 )
 def test_convbias():
@@ -304,6 +308,27 @@ def test_convbias():
     check_pygraph_dump(fwd, [inp, weight, bias], [result])
 
 
+@pytest.mark.skip(reason="does not support int4 when cuda version is lower than 10.2")
+def test_conv_bias_int4():
+    @trace(symbolic=True, capture_as_const=True)
+    def fwd(inp, weight, bias):
+        return F.quantized.conv_bias_activation(
+            inp,
+            weight,
+            bias,
+            dtype=dtype.quint4(scale=1.0, zero_point=0),
+            nonlinear_mode="relu",
+        )
+
+    inp = Tensor(
+        np.random.random((1, 3, 64, 64)), dtype=dtype.quint4(scale=1.0, zero_point=0)
+    )
+    weight = Tensor(np.random.random((32, 3, 3, 3)), dtype=dtype.qint4(scale=1.0))
+    bias = Tensor(np.random.random((1, 32, 1, 1)), dtype=dtype.qint32(scale=1.0))
+    result = fwd(inp, weight, bias)
+    check_pygraph_dump(fwd, [inp, weight, bias], [result])
+
+
 def test_batch_convbias():
     if is_cuda_available():
         return

imperative/src/impl/proxy_graph/mini_graph.h

@@ -616,6 +616,7 @@ public:
             descs.emplace_back();
             auto& desc = descs.back();
             desc.layout.dtype = minigraph.output_var(i)->dtype();
+            desc.layout.format = minigraph.output_var(i)->format();
             desc.comp_node = minigraph.output_var(i)->comp_node();
             if (auto* shape = sess.infer_shape(i, false)) {
                 desc.layout.init_contiguous_stride(*shape);

src/core/impl/tensor.cpp

@@ -435,6 +435,11 @@ DEF(DType dtype) : m_layout{dtype} {}
 DEF(CompNode node, DType dtype) : m_storage{node}, m_layout{dtype} {}
 
 //! allocate contiguous from given comp node, shape and dtype
+DEF(CompNode node, const TensorShape& shape, DType dtype)
+        : m_storage{node}, m_layout{dtype} {
+    resize(shape);
+}
+
 DEF(CompNode node, const TensorShape& shape, DType dtype, TensorFormat format)
         : m_storage{node}, m_layout{dtype, format} {
     resize(shape);

src/core/include/megbrain/tensor.h

@@ -339,8 +339,10 @@ public:
 
     //! allocate contiguous tensor
     MGE_WIN_DECLSPEC_FUC TensorND(
-            CompNode node, const TensorShape& shape, DType dtype = dtype::Float32{},
-            TensorFormat format = {});
+            CompNode node, const TensorShape& shape, DType dtype = dtype::Float32{});
+
+    MGE_WIN_DECLSPEC_FUC TensorND(
+            CompNode node, const TensorShape& shape, DType dtype, TensorFormat format);
 
     //! allocate contiguous tensor from given comp node and layout; layout
     //! is required to be contiguous, and its dtype and format would be used