Skip to content

Enable Float4_e2m1fn_x2 to Float4_e2m1fn_x2 copy #2310

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

Merged
merged 2 commits into from
Nov 13, 2025
+30 −2
Merged

Enable Float4_e2m1fn_x2 to Float4_e2m1fn_x2 copy #2310

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
44b32df
Enable Float4_e2m1fn_x2 to Float4_e2m1fn_x2 copy
CuiYifeng Nov 3, 2025
48cd4bb
Add UT for FP4 Copy
CuiYifeng Nov 7, 2025
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
12 changes: 12 additions & 0 deletions src/ATen/native/xpu/sycl/CopyKernel.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,6 +111,16 @@ void float8_copy_kernel_xpu(TensorIteratorBase& iter) {
}
}

void float4_copy_kernel_xpu(TensorIteratorBase& iter) {
ScalarType src_dtype = iter.dtype(1);

if (src_dtype == kFloat4_e2m1fn_x2) {
gpu_kernel_nocast(iter, CopyScalarFunc<Float4_e2m1fn_x2>());
} else {
TORCH_CHECK(false, "Copy from ", src_dtype, " to Float4_e2m1fn_x2 has not been supported.");
}
}

void copy_kernel(TensorIteratorBase& iter) {
ScalarType dtype = iter.common_dtype();
if (isQIntType(dtype)) {
Expand All @@ -119,6 +129,8 @@ void copy_kernel(TensorIteratorBase& iter) {
});
} else if (isFloat8Type(iter.dtype(0))) {
float8_copy_kernel_xpu(iter);
} else if (iter.dtype(0) == kFloat4_e2m1fn_x2) {
float4_copy_kernel_xpu(iter);
} else {
AT_DISPATCH_V2(
dtype,
Expand Down
20 changes: 18 additions & 2 deletions test/regressions/test_copy.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,8 +14,8 @@
class TestSimpleCopy(TestCase):
@dtypes(*float8_types_and(torch.float8_e8m0fnu, torch.float32))
def test_copy_and_clone(self, dtype):
a_cpu = torch.randn(16, 64, 28, 28)
b_cpu = torch.randn(16, 64, 28, 28)
a_cpu = torch.randn(16, 64, 28, 28).to(dtype)
b_cpu = torch.randn(16, 64, 28, 28).to(dtype)
a_xpu = a_cpu.to(xpu_device)
b_xpu = b_cpu.to(xpu_device)
# naive
Expand All @@ -27,6 +27,22 @@ def test_copy_and_clone(self, dtype):
b_xpu = a_xpu.clone(memory_format=torch.channels_last)
self.assertEqual(b_cpu, b_xpu.to(cpu_device))

def test_copy_and_clone_float4(self):
# Float4_e2m1fn_x2 copy is not implemented by CPU
a_cpu = torch.randn(16, 64, 28, 28).to(torch.uint8)
b_cpu = torch.randn(16, 64, 28, 28).to(torch.uint8)
a_xpu = a_cpu.to(xpu_device).view(torch.float4_e2m1fn_x2)
b_xpu = b_cpu.to(xpu_device).view(torch.float4_e2m1fn_x2)

b_cpu.copy_(a_cpu)
b_xpu.copy_(a_xpu)
# Float4_e2m1fn_x2 compare is not implemented CPU
self.assertEqual(b_cpu, b_xpu.view(torch.uint8).to(cpu_device))

b_cpu = a_cpu.clone(memory_format=torch.channels_last)
b_xpu = a_xpu.clone(memory_format=torch.channels_last)
self.assertEqual(b_cpu, b_xpu.view(torch.uint8).to(cpu_device))


instantiate_device_type_tests(TestSimpleCopy, globals(), only_for="xpu", allow_xpu=True)

Expand Down