Add bitwise and compare ops #47

xytintel · 2024-03-15T08:32:28Z

e.g. isnan, clamp, clamp_min, clamp_max, bitwise_and, bitwise_or, bitwise_xor, bitwise_not, _local_scalar_dense

fengyuan14 · 2024-03-25T01:08:02Z

src/aten/XPUScalar.cpp

+        q.memcpy(
+            (void*)value.const_data_ptr<scalar_t>(),
+            self.const_data_ptr<scalar_t>(),
+            sizeof(scalar_t));


Need synchronize the copy, or the value on host could not be valid.

src/aten/CompareOps.cpp

src/aten/sycl/CompareKernels.cpp

fengyuan14 · 2024-03-25T02:09:38Z

src/aten/sycl/CompareKernels.cpp

+
+template <typename opmath_t>
+struct ClampMaxFunctor {
+  opmath_t operator()(opmath_t v) const {


src/aten/sycl/CompareKernels.cpp

fengyuan14 · 2024-03-25T02:13:51Z

src/aten/sycl/CompareKernels.cpp

+  AT_DISPATCH_ALL_TYPES_AND2(
+      kHalf, kBFloat16, iter.dtype(), "clamp_min_xpu", [&]() {
+        auto lower = min_value.to<scalar_t>();
+        gpu_kernel(iter, ClampMinFunctor<scalar_t>(lower));


ditto, checking opmath.

fengyuan14 · 2024-03-25T02:14:01Z

src/aten/sycl/CompareKernels.cpp

+  AT_DISPATCH_ALL_TYPES_AND2(
+      kHalf, kBFloat16, iter.dtype(), "clamp_max_xpu", [&]() {
+        auto upper = max_value.to<scalar_t>();
+        gpu_kernel(iter, ClampMaxFunctor<scalar_t>(upper));


ditto, checking opmath.

xytintel · 2024-03-26T08:54:20Z

split pr

xytintel and others added 16 commits March 15, 2024 16:15

Create BinaryBitwiseOpsKernels.cpp

ecf4185

Create BinaryBitwiseOpsKernels.h

1623a8a

Update CompareKernels.cpp

8338eb3

Update CompareKernels.h

016b802

Create BinaryBitwiseOps.cpp

b36b28a

Update CompareOps.cpp

4fdb06b

Create XPUScalar.cpp

f361fd4

Update xpu_functions.yaml

48b1675

Update test_compare.py

3b3f797

Merge branch 'main' into xyt/bitwise

a934e1c

Update CompareKernels.cpp

114296a

Update CompareKernels.h

809ac03

Update CompareOps.cpp

dde92a8

Update xpu_functions.yaml

de96c97

Merge branch 'main' into xyt/bitwise

8c37fdd

Merge branch 'main' into xyt/bitwise

8b95a43