initial cpu and gpu testing with pad

src/targets/cpu/lowering.cpp

@@ -309,7 +309,7 @@ struct cpu_pad
         result.visit([&](auto output) { std::fill(output.begin(), output.end(), op.value); });
 
         visit_all(result, args[0])([&](auto output, auto input) {
-            shape_for_each(output.get_shape(), [&](const auto& idx) {
+            shape_for_each(input.get_shape(), [&](const auto& idx) {
                 std::vector<std::size_t> new_idx(idx.size());
                 std::transform(
                     idx.begin(), idx.end(), op.pads.begin(), new_idx.begin(), [](auto i, auto j) {

src/targets/gpu/CMakeLists.txt

@@ -28,6 +28,7 @@ add_library(migraphx_device
     device/contiguous.cpp
     device/mul.cpp
     device/concat.cpp
+    device/pad.cpp
 )
 set_target_properties(migraphx_device PROPERTIES EXPORT_NAME device)
 rocm_clang_tidy_check(migraphx_device)
@@ -56,6 +57,7 @@ add_library(migraphx_gpu
     sigmoid.cpp
     abs.cpp
     elu.cpp
+    pad.cpp
 )
 set_target_properties(migraphx_gpu PROPERTIES EXPORT_NAME gpu)
 rocm_clang_tidy_check(migraphx_gpu)

src/targets/gpu/device/include/migraphx/gpu/device/nary.hpp

@@ -313,6 +313,12 @@ void nary_impl(hipStream_t stream, F f, argument result, Arguments... args)
         nary_nonstandard_impl(stream, f, result, args...);
 }
 
+template <class F>
+void nary_impl(hipStream_t stream, F f, argument result)
+{
+    nary_standard_impl(stream, f, result);
+}
+
 template <class... Arguments>
 auto nary_nonstandard(hipStream_t stream, argument result, Arguments... args)
 {

src/targets/gpu/device/pad.cpp

+#include <migraphx/shape.hpp>
+#include <migraphx/argument.hpp>
+#include <migraphx/gpu/device/nary.hpp>
+#include <migraphx/gpu/device/pad.hpp>
+#include <migraphx/gpu/device/tensor.hpp>
+#include <migraphx/gpu/device/launch.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+namespace device {
+
+argument pad(hipStream_t stream,
+                argument result,
+                argument arg1,
+                float value,
+                std::vector<std::int64_t> pads)
+{
+    std::size_t nelements = arg1.get_shape().elements();
+
+    nary(stream, result)([=] { return value; });
+    visit_all(result, arg1)([&](auto output, auto input) {
+        visit_tensor_size(result.get_shape().lens().size(), [&](auto ndim) {
+            std::size_t offsets[ndim];
+            std::copy(pads.begin(), pads.begin() + ndim, offsets);
+            auto* outptr      = output.data();
+            const auto* inptr = input.data();
+            hip_tensor_descriptor<ndim> desc_input(input.get_shape());
+            hip_tensor_descriptor<ndim> desc_output(output.get_shape());
+            gs_launch(stream, nelements)(
+                [=](auto i) { 
+                    auto idx = desc_input.multi(i);
+                    for(std::size_t j = 0; j < ndim;j++) {
+                        idx[j] += offsets[j];
+                    }
+                    outptr[desc_output.linear(idx)] = inptr[i]; 
+                });
+        });
+    });
+    return result;
+}
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx

src/targets/gpu/include/migraphx/gpu/device/pad.hpp

+
+#ifndef MIGRAPHX_GUARD_RTGLIB_DEVICE_PAD_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DEVICE_PAD_HPP
+
+#include <migraphx/argument.hpp>
+#include <migraphx/config.hpp>
+#include <hip/hip_runtime_api.h>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+namespace device {
+
+argument pad(hipStream_t stream,
+                argument result,
+                argument arg1,
+                float value,
+                std::vector<std::int64_t> pads);
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/targets/gpu/include/migraphx/gpu/pad.hpp

+#ifndef MIGRAPHX_GUARD_RTGLIB_PAD_HPP
+#define MIGRAPHX_GUARD_RTGLIB_PAD_HPP
+
+#include <migraphx/gpu/lowering.hpp>
+#include <migraphx/manage_ptr.hpp>
+#include <migraphx/instruction.hpp>
+#include <migraphx/operators.hpp>
+#include <migraphx/generate.hpp>
+#include <migraphx/shape_for_each.hpp>
+#include <migraphx/config.hpp>
+#include <migraphx/gpu/miopen.hpp>
+#include <migraphx/gpu/hip.hpp>
+#include <migraphx/dfor.hpp>
+#include <migraphx/gpu/device/pad.hpp>
+#include <migraphx/gpu/device/add.hpp>
+#include <migraphx/iterator_for.hpp>
+#include <migraphx/gpu/rocblas.hpp>
+#include <migraphx/gpu/context.hpp>
+#include <utility>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+
+struct hip_pad
+{
+    op::pad op;
+
+    std::string name() const { return "gpu::pad"; }
+    shape compute_shape(std::vector<shape> inputs) const;
+    argument
+    compute(context& ctx, const shape& output_shape, const std::vector<argument>& args) const;
+    int output_alias(const std::vector<shape>& shapes) const { return shapes.size() - 1; }
+};
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/targets/gpu/lowering.cpp

@@ -40,6 +40,7 @@
 #include <migraphx/gpu/pooling.hpp>
 #include <migraphx/gpu/gemm.hpp>
 #include <migraphx/gpu/concat.hpp>
+#include <migraphx/gpu/pad.hpp>
 #include <utility>
 #include <functional>
 #include <algorithm>
@@ -90,6 +91,7 @@ struct miopen_apply
         add_extend_op<miopen_contiguous, op::contiguous>("contiguous");
         add_extend_op<hip_concat, op::concat>("concat");
         add_extend_op<miopen_softmax, op::softmax>("softmax");
+        add_extend_op<hip_pad, op::pad>("pad");
 
         add_convolution_op();
         add_pooling_op();

src/targets/gpu/pad.cpp

+#include <migraphx/gpu/pad.hpp>
+#include <migraphx/operators.hpp>
+#include <migraphx/manage_ptr.hpp>
+#include <migraphx/gpu/miopen.hpp>
+#include <migraphx/gpu/device/pad.hpp>
+#include <utility>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+
+shape hip_pad::compute_shape(std::vector<shape> inputs) const
+{
+    inputs.pop_back();
+    return op.compute_shape(inputs);
+}
+
+argument hip_pad::compute(context& ctx,
+                             const shape&,
+                             const std::vector<argument>& args) const
+{
+    return device::pad(ctx.get_stream().get(), args.back(), args.front(), op.value, op.pads);
+}
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx

test/cpu_ops_test.cpp

@@ -1283,4 +1283,21 @@ TEST_CASE(min_test)
     EXPECT(migraphx::verify_range(results_vector, gold));
 }
 
+TEST_CASE(pad_test)
+{
+    migraphx::program p;
+    migraphx::shape s{migraphx::shape::float_type, {2,2}};
+    auto l0 = p.add_literal(migraphx::literal{s, {1, 2, 3, 4}});
+    p.add_instruction(migraphx::op::pad{{1,1,1,1}}, l0);
+    p.compile(migraphx::cpu::target{});
+    auto result = p.eval({});
+    std::vector<float> results_vector(16);
+    result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
+    std::vector<float> gold { 0, 0, 0, 0,
+                             0, 1, 2, 0,
+                             0, 3, 4, 0,
+                             0, 0, 0, 0};
+    EXPECT(migraphx::verify_range(results_vector, gold));
+}
+
 int main(int argc, const char* argv[]) { test::run(argc, argv); }

test/gpu/miopen.cpp

@@ -934,6 +934,19 @@ struct test_concat_relu
     }
 };
 
+struct test_pad
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        migraphx::shape s0{migraphx::shape::int32_type, {1, 96, 165, 165}};
+        std::vector<int64_t> pads = {0,0,0,0,0,0,1,1};
+        auto l0 = p.add_parameter("x", s0);
+        p.add_instruction(migraphx::op::pad{pads}, l0);
+        return p;
+    }
+};
+
 void manual_identity()
 {
     migraphx::program p;
@@ -1023,6 +1036,7 @@ int main()
     verify_program<test_concat>();
     verify_program<test_concat2>();
     verify_program<test_concat_relu>();
+    verify_program<test_pad>();
     verify_program<test_add>();
     verify_program<test_add_half>();
     verify_program<test_mul>();