Merge branch 'develop' into blas_tuning

src/propagate_constant.cpp

@@ -27,11 +27,14 @@
 #include <migraphx/literal.hpp>
 #include <migraphx/functional.hpp>
 #include <migraphx/par_for.hpp>
+#include <migraphx/env.hpp>
 #include <unordered_set>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 
+MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_TRACE_PROPAGATE_CONSTANT)
+
 bool skip_propogate(instruction_ref ins)
 {
     if(ins->name() == "contiguous")
@@ -85,6 +88,19 @@ void propagate_constant::apply(module& m) const
     {
         if(not literals[i].empty())
         {
+            if(enabled(MIGRAPHX_TRACE_PROPAGATE_CONSTANT{}))
+            {
+                std::cout << "Constant replace: " << std::endl;
+                std::vector<instruction_ref> inss;
+                fix([&](auto self, auto ins) {
+                    if(contains(inss, ins))
+                        return;
+                    for(auto input : ins->inputs())
+                        self(input);
+                    inss.push_back(ins);
+                })(const_instrs_vec[i]);
+                m.debug_print(inss);
+            }
             assert(literals[i].get_shape() == const_instrs_vec[i]->get_shape());
             auto l = m.add_literal(literals[i].get_shape(), literals[i].data());
             m.replace_instruction(const_instrs_vec[i], l);

src/py/migraphx_py.cpp

@@ -62,6 +62,7 @@ namespace py = pybind11;
     PYBIND11_MODULE(__VA_ARGS__)      \
     MIGRAPHX_POP_WARNING
 
+#define MIGRAPHX_PYTHON_GENERATE_SHAPE_ENUM(x, t) .value(#x, migraphx::shape::type_t::x)
 namespace migraphx {
 
 migraphx::value to_value(py::kwargs kwargs);
@@ -235,7 +236,8 @@ migraphx::shape to_shape(const py::buffer_info& info)
 
 MIGRAPHX_PYBIND11_MODULE(migraphx, m)
 {
-    py::class_<migraphx::shape>(m, "shape")
+    py::class_<migraphx::shape> shape_cls(m, "shape");
+    shape_cls
         .def(py::init([](py::kwargs kwargs) {
             auto v    = migraphx::to_value(kwargs);
             auto t    = migraphx::shape::parse_type(v.get("type", "float"));
@@ -261,6 +263,9 @@ MIGRAPHX_PYBIND11_MODULE(migraphx, m)
         .def("__ne__", std::not_equal_to<migraphx::shape>{})
         .def("__repr__", [](const migraphx::shape& s) { return migraphx::to_string(s); });
 
+    py::enum_<migraphx::shape::type_t>(shape_cls, "type_t")
+        MIGRAPHX_SHAPE_VISIT_TYPES(MIGRAPHX_PYTHON_GENERATE_SHAPE_ENUM);
+
     py::class_<migraphx::argument>(m, "argument", py::buffer_protocol())
         .def_buffer([](migraphx::argument& x) -> py::buffer_info { return to_buffer_info(x); })
         .def(py::init([](py::buffer b) {

src/shape.cpp

@@ -481,6 +481,15 @@ shape shape::with_type(type_t t) const
 
 shape shape::to_dynamic() const
 {
+    if(not sub_shapes().empty())
+    {
+        std::vector<shape> subs;
+        std::transform(sub_shapes().cbegin(),
+                       sub_shapes().cend(),
+                       std::back_inserter(subs),
+                       [](auto s) { return s.to_dynamic(); });
+        return {subs};
+    }
     if(this->dynamic())
     {
         return *this;
@@ -488,6 +497,30 @@ shape shape::to_dynamic() const
     return {type(), lens(), lens(), {}};
 }
 
+shape shape::to_static(std::size_t x) const
+{
+    if(not sub_shapes().empty())
+    {
+        std::vector<shape> subs;
+        std::transform(sub_shapes().cbegin(),
+                       sub_shapes().cend(),
+                       std::back_inserter(subs),
+                       [&](auto s) { return s.to_static(x); });
+        return {subs};
+    }
+    if(not this->dynamic())
+    {
+        return *this;
+    }
+    auto static_lens = this->max_lens();
+    std::transform(static_lens.begin(),
+                   static_lens.end(),
+                   this->dyn_dims().cbegin(),
+                   static_lens.begin(),
+                   [&](auto sl, auto dd) { return dd.is_fixed() ? sl : x; });
+    return {type(), static_lens};
+}
+
 std::size_t shape::element_space() const { return impl->element_space(); }
 
 std::string shape::type_string() const { return name(this->type()); }

src/simplify_algebra.cpp

@@ -52,8 +52,9 @@ auto op_lit_broadcast(std::string op, std::string x, std::string y)
 
 auto conv_const_weights()
 {
-    return match::name("convolution")(match::used_once(),
-                                      match::args(match::any(), match::is_constant().bind("w")));
+    return match::name("convolution")(
+        match::used_once(),
+        match::args(match::none_of(match::is_constant()), match::is_constant().bind("w")));
 }
 
 auto reduction() { return match::name_contains("reduce"); }
@@ -203,7 +204,12 @@ struct find_mul_slice_conv
     }
 };
 
-// a * (x + b) => a * x + a * b
+// ******************************
+//  a * (x + b) => a * x + a * b
+// ******************************
+// When a * (x + b) is followed by another add of constant, then the
+// additional add can be const folded. Also, better fusions can be applied
+// when the add comes after.
 struct find_mul_add
 {
     auto matcher() const
@@ -268,6 +274,32 @@ struct find_dot_add
     }
 };
 
+struct find_conv_add
+{
+    auto matcher() const
+    {
+        auto add = match::name("add")(
+            match::either_arg(0, 1)(match::any().bind("x"),
+                                    match::any_of(match::is_constant()).bind("a")),
+            match::used_once());
+        return match::name("convolution")(match::used_once(),
+                                          match::args(add, match::is_constant().bind("w")));
+    }
+
+    void apply(module& m, const match::matcher_result& r) const
+    {
+        auto ins   = r.result;
+        auto a_ins = r.instructions["a"];
+        auto x_ins = r.instructions["x"];
+        auto w_ins = r.instructions["w"];
+
+        auto conv1 = m.insert_instruction(ins, ins->get_operator(), a_ins, w_ins);
+        auto conv2 = m.insert_instruction(ins, ins->get_operator(), x_ins, w_ins);
+
+        m.replace_instruction(ins, make_op("add"), conv1, conv2);
+    }
+};
+
 struct find_add_lit_broadcast
 {
     auto matcher() const
@@ -1239,6 +1271,7 @@ void simplify_algebra::apply(module& m) const
                             find_neg_unit_ops{},
                             find_zero_ops{},
                             find_dot_add{},
+                            find_conv_add{},
                             find_div_const{},
                             find_sub_const{},
                             find_rsqrt{},

src/split_single_dyn_dim.cpp

@@ -28,6 +28,7 @@
 #include <migraphx/functional.hpp>
 #include <migraphx/make_op.hpp>
 #include <migraphx/ranges.hpp>
+#include <migraphx/matcher.hpp>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
@@ -67,6 +68,37 @@ has_one_dyn_dim(const std::unordered_map<std::string, shape>& param_shapes)
                                     dds_it->max};
 }
 
+namespace {
+struct find_static_2in_broadcasts
+{
+    // Convert 2 input static shape broadcast/multibroadcast into 1 input version.
+    // Some compiler passes (ex. simplify_algebra) only support the 1 input versions
+    // of the broadcasting operators.
+    auto matcher() const
+    {
+        return match::broadcast(match::nargs(2),
+                                match::arg(0)(match::static_shape()),
+                                match::arg(1)(match::static_shape()));
+    }
+
+    void apply(module& m, const match::matcher_result& mr) const
+    {
+        auto ins          = mr.result;
+        auto out_lens     = ins->get_shape().lens();
+        auto broadcast_op = ins->get_operator();
+        if(broadcast_op.name() == "broadcast")
+        {
+            broadcast_op.from_value({{"out_lens", out_lens}});
+        }
+        else
+        {
+            broadcast_op.from_value({{"out_lens", out_lens}, {"out_dyn_dims", {}}});
+        }
+        m.replace_instruction(ins, broadcast_op, ins->inputs().at(0));
+    }
+};
+} // namespace
+
 /**
  * Makes all the shapes in the dynamic_dimension range.
  * Probably won't work for `if` and `loop` instructions, depending on how the submodules for those
@@ -97,6 +129,7 @@ void split_single_dyn_dim::apply(module_pass_manager& mpm) const
                 dd_check->dyn_param_str, migraphx::shape{dyn_param_shape.type(), static_lens});
             auto outputs = submod->add_instructions(mm, map_ins);
             submod->add_return({outputs});
+            match::find_matches(*submod, find_static_2in_broadcasts{});
             submodules.push_back(submod);
         }
         // redirect to select_module operator and return

src/targets/cpu/target.cpp

@@ -82,7 +82,6 @@ std::vector<pass> target::get_passes(migraphx::context& gctx, const compile_opti
             dead_code_elimination{},
             simplify_algebra{},
             simplify_reshapes{},
-            layout_nhwc{},
             dead_code_elimination{},
             simplify_reshapes{},
             simplify_algebra{},

src/targets/gpu/CMakeLists.txt

@@ -33,7 +33,11 @@ if(NOT TARGET MIOpen)
     message(SEND_ERROR "Cant find miopen")
 endif()
 
-set(MIGRAPHX_USE_HIPRTC OFF CACHE BOOL "Use hipRTC APIs")
+if(BUILD_DEV)
+    set(MIGRAPHX_USE_HIPRTC OFF CACHE BOOL "Use hipRTC APIs")
+else()
+    set(MIGRAPHX_USE_HIPRTC ON CACHE BOOL "Use hipRTC APIs")
+endif()
 
 include(Embed)
 file(GLOB KERNEL_FILES ${CONFIGURE_DEPENDS}

src/targets/gpu/compile_gen.cpp

@@ -168,7 +168,7 @@ std::string make_transformer_args(std::vector<std::string> transformers)
     return join_strings(std::move(transformers), ", ");
 }
 
-std::string generate_pointwise(const module& pm, const std::string& name)
+void generate_pointwise(cpp_generator& gg, const module& pm, const std::string& name)
 {
     module m = pm;
     run_passes(m, {eliminate_common_subexpression{}, dead_code_elimination{}});
@@ -184,8 +184,131 @@ std::string generate_pointwise(const module& pm, const std::string& name)
     // Add explict conversions
     g.fresult(
         [](const shape& s) { return "migraphx::convert<" + shape::cpp_type(s.type()) + ">"; });
-    g.create_function(
-        g.generate_module(m).set_attributes({"__device__"}).set_generic_types(m).set_name(name));
+    gg.create_function(g.generate_module(m)
+                           .set_attributes({"__device__", "__attribute__((const))"})
+                           .set_generic_types(m)
+                           .set_name(name));
+}
+std::string generate_pointwise(const module& pm, const std::string& name)
+{
+    cpp_generator g;
+    generate_pointwise(g, pm, name);
+    return g.str();
+}
+
+std::string reduce_op::str() const
+{
+    return write + "(r.reduce(" + reduction + ", " + init + ", " + read + ")(" + input + "))";
+}
+void reduce_op::set(instruction_ref ins, const operation& op)
+{
+    if(op.name() == "reduce_sum")
+    {
+        reduction = "op::sum{}";
+    }
+    else if(op.name() == "reduce_mean")
+    {
+        auto s               = ins->inputs().front()->get_shape();
+        auto reduce_elements = s.elements() / ins->get_shape().elements();
+        auto reduce_type     = s.type();
+        reduction            = "op::sum{}";
+        std::string mean     = "op::mean<" + std::to_string(reduce_elements) + ">{}";
+        // Use float accumulator when reduction size is too large for half
+        if(reduce_type == shape::half_type and reduce_elements > 16384)
+            read = "compose(" + mean + ", op::convert_to<float>{})";
+        else if(contains({shape::float_type, shape::half_type, shape::double_type}, reduce_type))
+            read = mean;
+        else
+            write = mean;
+    }
+    else if(op.name() == "reduce_max")
+    {
+        reduction = "op::max{}";
+        init      = "lowest{}";
+    }
+    else if(op.name() == "reduce_min")
+    {
+        reduction = "op::min{}";
+        init      = "highest{}";
+    }
+    else if(op.name() == "reduce_prod")
+    {
+        reduction = "op::product{}";
+        init      = "1";
+    }
+    else
+    {
+        MIGRAPHX_THROW("Unsupported reduce");
+    }
+}
+std::string reduce_op::generate(instruction_ref ins, const std::string& x)
+{
+    reduce_op r{x};
+    r.set(ins, ins->get_operator());
+    return r.str();
+}
+
+static bool use_lazy_inner(instruction_ref ins)
+{
+    if(ins->outputs().size() != 1)
+        return false;
+    auto output = ins->outputs().front();
+    return contains(output->name(), "reduce") or output->name() == "@return";
+}
+
+std::string generate_reduce(const module& m, const std::string& name)
+{
+    cpp_generator g;
+    auto ilens    = m.get_parameter_shapes().begin()->second.lens();
+    std::size_t i = 0;
+    auto f        = g.generate_module(m, [&](instruction_ref ins, const auto& names) {
+        if(contains(ins->name(), "reduce"))
+        {
+            return reduce_op::generate(ins, names.at(ins->inputs().front()));
+        }
+        else if(ins->name() == "pointwise")
+        {
+            auto pointwise_name = "pointwise" + std::to_string(i);
+            i++;
+            generate_pointwise(g, *ins->module_inputs().front(), pointwise_name);
+            std::vector<instruction_ref> tensors;
+            std::copy_if(ins->inputs().begin(),
+                         ins->inputs().end(),
+                         std::back_inserter(tensors),
+                         [&](auto input) {
+                             return input->get_shape().lens() == ilens and
+                                    not input->get_shape().broadcasted();
+                         });
+            auto inner_names = names;
+            for(auto input : tensors)
+                inner_names[input] += "_lambda_param";
+            auto call_function =
+                pointwise_name + "(" +
+                join_strings(cpp_generator::to_args(ins->inputs(), inner_names), ", ") + ")";
+            if(tensors.empty())
+                return call_function;
+            const std::string inner_template =
+                "r.${inner}([=](${params}) { return ${call}; })(${args})";
+            std::string inner_name = use_lazy_inner(ins) ? "lazy_inner" : "inner";
+            auto args              = cpp_generator::to_args(tensors, names);
+            auto params            = cpp_generator::to_args(tensors, inner_names);
+            std::transform(
+                params.begin(), params.end(), params.begin(), [](auto s) { return "auto " + s; });
+            return interpolate_string(inner_template,
+                                      {{"inner", inner_name},
+                                       {"params", join_strings(params, ", ")},
+                                       {"args", join_strings(args, ", ")},
+                                       {"call", call_function}});
+        }
+        else if(ins->name() == "multibroadcast")
+        {
+            return names.at(ins->inputs().front());
+        }
+        MIGRAPHX_THROW("Unknown operator: " + ins->name());
+    });
+    f.set_attributes({"__device__", "__attribute__((const))"}).set_generic_types(m).set_name(name);
+    f.add_generic_param("r");
+    g.create_function(f);
     return g.str();
 }
 
@@ -196,7 +319,17 @@ static std::vector<std::string> get_op_names(const module& m)
     {
         if(starts_with(ins.name(), "@"))
             continue;
-        result.push_back(ins.name());
+        if(ins.name() == "multibroadcast")
+            continue;
+        if(ins.name() == "pointwise")
+        {
+            auto names = get_op_names(*ins.module_inputs().front());
+            result.insert(result.end(), names.begin(), names.end());
+        }
+        else
+        {
+            result.push_back(ins.name());
+        }
     }
     return result;
 }

src/targets/gpu/driver/CMakeLists.txt

@@ -26,5 +26,6 @@ file(GLOB GPU_DRIVER_SRCS ${CONFIGURE_DEPENDS} ${CMAKE_CURRENT_SOURCE_DIR}/*.cpp
 add_executable(gpu-driver
     ${GPU_DRIVER_SRCS}
 )
+rocm_clang_tidy_check(gpu-driver)
 target_include_directories(gpu-driver PRIVATE include)
 target_link_libraries(gpu-driver PRIVATE migraphx_gpu)

src/targets/gpu/driver/include/migraphx/gpu/driver/action.hpp

@@ -44,7 +44,7 @@ struct auto_register_action
     template <class T>
     static void apply()
     {
-        auto name = get_type_name<T>();
+        const auto& name = get_type_name<T>();
         register_action(name.substr(name.rfind("::") + 2),
                         [](auto&&... xs) { T::apply(std::forward<decltype(xs)>(xs)...); });
     }

src/targets/gpu/hip.cpp

@@ -189,8 +189,20 @@ argument register_on_gpu(const argument& arg)
 
 argument to_gpu(const argument& arg, bool host)
 {
-    auto p = write_to_gpu(arg.data(), arg.get_shape().bytes(), host);
-    return {arg.get_shape(), p};
+    argument result;
+    arg.visit(
+        [&](auto x) {
+            auto p = write_to_gpu(arg.data(), arg.get_shape().bytes(), host);
+            result = {x.get_shape(), p};
+        },
+        [&](const auto& xs) {
+            std::vector<argument> args;
+            std::transform(xs.begin(), xs.end(), std::back_inserter(args), [&](auto x) {
+                return to_gpu(x, host);
+            });
+            result = argument{args};
+        });
+    return result;
 }
 
 argument from_gpu(const argument& arg)

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

@@ -26,6 +26,7 @@
 
 #include <migraphx/config.hpp>
 #include <migraphx/module_ref.hpp>
+#include <migraphx/instruction_ref.hpp>
 #include <string>
 #include <unordered_map>
 #include <vector>
@@ -34,6 +35,7 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 
 struct shape;
+struct operation;
 
 namespace gpu {
 
@@ -72,8 +74,23 @@ std::string make_transformer_args(Ts... xs)
 
 std::string generate_pointwise(const module& pm, const std::string& name);
 
+std::string generate_reduce(const module& m, const std::string& name);
+
 std::string generate_name_from_ops(const module& m);
 
+struct reduce_op
+{
+    std::string input     = "";
+    std::string reduction = "";
+    std::string init      = "0";
+    std::string read      = "op::id{}";
+    std::string write     = "op::id{}";
+
+    void set(instruction_ref ins, const operation& op);
+    std::string str() const;
+    static std::string generate(instruction_ref ins, const std::string& x);
+};
+
 } // namespace gen
 } // namespace gpu
 } // namespace MIGRAPHX_INLINE_NS

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

@@ -71,6 +71,8 @@ operation compile_hip_code_object(const std::string& content, hip_compile_option
 
 std::size_t compute_block_size(std::size_t n, std::size_t max_block_size = 1024);
 
+std::string generate_make_shape(const shape& s);
+
 } // namespace gpu
 } // namespace MIGRAPHX_INLINE_NS
 } // namespace migraphx

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

@@ -21,8 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  * THE SOFTWARE.
  */
-#ifndef MIGRAPHX_GUARD_RTGLIB_CONVOLUTION_HPP
-#define MIGRAPHX_GUARD_RTGLIB_CONVOLUTION_HPP
+#ifndef MIGRAPHX_GUARD_RTGLIB_GPU_CONVOLUTION_HPP
+#define MIGRAPHX_GUARD_RTGLIB_GPU_CONVOLUTION_HPP
 
 #include <migraphx/shape.hpp>
 #include <migraphx/generate.hpp>

src/targets/gpu/jit/reduce.cpp

@@ -60,15 +60,6 @@ __global__ void reduce_kernel(void* input_p, void* output_p)
 
 )__migraphx__";
 
-static std::size_t get_reduce_elements(const std::vector<shape>& inputs)
-{
-    return inputs.front().elements() / inputs.back().elements();
-}
-static std::size_t get_reduce_elements(const std::vector<instruction_ref>& inputs)
-{
-    return get_reduce_elements(to_shapes(inputs));
-}
-
 static std::vector<std::size_t> get_reduce_lens(const std::vector<std::size_t>& input_lens,
                                                 const std::vector<std::size_t>& output_lens)
 {
@@ -86,9 +77,28 @@ static std::vector<std::size_t> get_reduce_lens(const std::vector<std::size_t>&
     return reduce_lens;
 }
 
-static std::string get_reduce_algo(const std::vector<shape>& inputs)
+template <class T>
+static shape get_reduced_shape(const shape& s, const std::vector<T>& axes)
+{
+    auto lens = s.lens();
+    std::fill(lens.begin(), lens.end(), 1);
+    for(const auto& axis : axes)
+        lens[axis] = s.lens()[axis];
+    return shape{s.type(), lens};
+}
+
+template <class T>
+static shape get_output_shape(const shape& s, const std::vector<T>& axes)
+{
+    auto lens = s.lens();
+    for(const auto& axis : axes)
+        lens[axis] = 1;
+    return shape{s.type(), lens};
+}
+
+template <class ReduceLens>
+static std::string get_reduce_algo(const std::vector<shape>& inputs, ReduceLens rlens)
 {
-    auto rlens      = get_reduce_lens(inputs.front().lens(), inputs.back().lens());
     const auto init = std::numeric_limits<std::size_t>::max();
     // The minimum stride
     auto min_stride = std::inner_product(
@@ -103,11 +113,27 @@ static std::string get_reduce_algo(const std::vector<shape>& inputs)
     return "block";
 }
 
-struct reduce_compiler : compiler<reduce_compiler>
+static std::string get_reduce_algo(const std::vector<shape>& inputs)
+{
+    auto rlens = get_reduce_lens(inputs.front().lens(), inputs.back().lens());
+    return get_reduce_algo(inputs, rlens);
+}
+
+struct simple_reduce_compiler : compiler<simple_reduce_compiler>
 {
     std::vector<std::string> names() const
     {
-        return {"reduce", "reduce_sum", "reduce_mean", "reduce_max", "reduce_min", "reduce_prod"};
+        return {"simple_reduce",
+                "reduce_sum",
+                "reduce_mean",
+                "reduce_max",
+                "reduce_min",
+                "reduce_prod"};
+    }
+
+    static std::size_t get_reduce_elements(const std::vector<shape>& inputs)
+    {
+        return inputs.front().elements() / inputs.back().elements();
     }
 
     operation compile_op(context& ctx, const std::vector<shape>& inputs, const value& v) const
@@ -157,44 +183,108 @@ struct reduce_compiler : compiler<reduce_compiler>
     compiler_replace compile(context& ctx, instruction_ref ins, const operation& op) const
     {
         value v = value::object{};
-        if(op.name() == "reduce_sum")
-        {
-            v["reduction"] = "op::sum{}";
-        }
-        else if(op.name() == "reduce_mean")
-        {
-            auto reduce_elements = get_reduce_elements(ins->inputs());
-            auto reduce_type     = ins->inputs().front()->get_shape().type();
-            v["reduction"]       = "op::sum{}";
-            std::string mean     = "op::mean<" + std::to_string(reduce_elements) + ">{}";
-            // Use float accumulator when reduction size is too large for half
-            if(reduce_type == shape::half_type and reduce_elements > 16384)
-                v["read"] = "compose(" + mean + ", op::convert_to<float>{})";
-            else if(contains({shape::float_type, shape::half_type, shape::double_type},
-                             reduce_type))
-                v["read"] = mean;
-            else
-                v["write"] = mean;
-        }
-        else if(op.name() == "reduce_max")
-        {
-            v["reduction"] = "op::max{}";
-            v["init"]      = "lowest{}";
-        }
-        else if(op.name() == "reduce_min")
+        reduce_op r{};
+        r.set(ins, op);
+        v["reduction"] = r.reduction;
+        v["read"]      = r.read;
+        v["write"]     = r.write;
+        v["init"]      = r.init;
+        return replace(compile_op(ctx, to_shapes(ins->inputs()), v));
+    }
+};
+
+static const char* const fused_reduce_kernel = R"__migraphx__(
+#include <migraphx/kernels/index.hpp>
+#include <migraphx/kernels/reduce.hpp>
+#include <migraphx/kernels/pointwise.hpp>
+#include <migraphx/kernels/vectorize.hpp>
+#include <args.hpp>
+
+namespace migraphx {
+
+${preamble}
+
+extern "C" {
+MIGRAPHX_GLOBAL void ${kernel}(${params})
+{
+    transform_args(make_tensors(), rotate_last(), ${transformers})(${args})([](auto y, auto... xs) {
+        fused_reduce<reduce::${algo}, ${reduced}>(y, partial(${lambda})(xs...));
+    });
+}
+    
+}
+
+} // namespace migraphx
+
+)__migraphx__";
+
+struct fused_reduce_compiler : compiler<fused_reduce_compiler>
+{
+    std::vector<std::string> names() const { return {"fused_reduce"}; }
+
+    operation compile_op(context& ctx, const std::vector<shape>& inputs, const value& v) const
+    {
+        auto axes           = v.at("axes").to_vector<std::size_t>();
+        auto virtual_inputs = inputs;
+        virtual_inputs.push_back(get_reduced_shape(inputs.front(), axes));
+        virtual_inputs.push_back(get_output_shape(inputs.front(), axes));
+        virtual_inputs           = reduce_dims(virtual_inputs);
+        auto reduce_output_shape = virtual_inputs.back();
+        virtual_inputs.pop_back();
+        auto reduction_shape = virtual_inputs.back();
+        virtual_inputs.pop_back();
+
+        hip_compile_options options;
+        options.inputs         = inputs;
+        options.output         = inputs.back();
+        options.virtual_inputs = virtual_inputs;
+        auto faxis             = find_fast_axis({options.virtual_inputs.front()});
+        vectorize vec{};
+        auto nelements = reduce_output_shape.elements();
+        auto algo = v.get("algo", get_reduce_algo(options.virtual_inputs, reduction_shape.lens()));
+        if(algo == "block")
         {
-            v["reduction"] = "op::min{}";
-            v["init"]      = "highest{}";
+            // Vectorize if the axis is a reduction axis
+            if(reduce_output_shape.lens()[faxis] == 1)
+                vec = vectorize::elements(ctx, faxis, options.virtual_inputs);
+            auto relements  = reduction_shape.elements() / vec.size;
+            auto block_size = compute_block_size(relements, 256);
+            if(relements >= block_size * 256)
+                algo = "block_large";
+            options.set_launch_params(
+                v, compute_global_for(ctx, nelements * block_size, 256), block_size);
         }
-        else if(op.name() == "reduce_prod")
+        else if(algo == "lane")
         {
-            v["reduction"] = "op::product{}";
-            v["init"]      = "1";
+            options.set_launch_params(v, compute_global_for(ctx, nelements, 256));
         }
         else
         {
-            MIGRAPHX_THROW("Unsupported reduce");
+            MIGRAPHX_THROW("Unknown reduce algo: " + algo);
         }
+        options.kernel_name = v.get("kernel", "reduce_kernel");
+        auto src            = interpolate_string(
+            fused_reduce_kernel,
+            {{"kernel", options.kernel_name},
+             {"params", enum_params(inputs.size(), "void * private_p")},
+             {"args", enum_params(inputs.size(), "private_p")},
+             {"algo", algo},
+             {"reduced", "decltype(" + generate_make_shape(reduce_output_shape) + ")"},
+             {"lambda", v.at("lambda").to<std::string>()},
+             {"transformers", make_transformer_args(vec)},
+             {"preamble", v.get("preamble", std::string{})}});
+        options.params += "-Wno-float-equal";
+        return compile_hip_code_object(src, options);
+    }
+
+    compiler_replace compile(context& ctx, instruction_ref ins, const operation& op) const
+    {
+        assert(not ins->module_inputs().empty());
+        auto v        = op.to_value();
+        auto* rm      = ins->module_inputs().front();
+        v["preamble"] = generate_reduce(*rm, "fused_reduce_op");
+        v["lambda"]   = "MIGRAPHX_LIFT(fused_reduce_op)";
+        v["kernel"]   = generate_name_from_ops(*rm) + "_kernel";
         return replace(compile_op(ctx, to_shapes(ins->inputs()), v));
     }
 };

src/targets/gpu/kernels/include/migraphx/kernels/functional.hpp

@@ -195,6 +195,14 @@ constexpr auto compose(Fs... fs)
     })(fs...);
 }
 
+template <class F>
+constexpr auto partial(F f)
+{
+    return [=](auto... xs) {
+        return [=](auto&&... ys) { return f(xs..., static_cast<decltype(ys)>(ys)...); };
+    };
+}
+
 template <class... Ts>
 constexpr auto pack(Ts... xs)
 {

src/targets/gpu/kernels/include/migraphx/kernels/index.hpp

@@ -233,6 +233,12 @@ struct index
     }
 };
 
+#ifdef MIGRAPHX_NLOCAL
+#define MIGRAPHX_GLOBAL \
+    __global__ __attribute__((amdgpu_flat_work_group_size(MIGRAPHX_NLOCAL, MIGRAPHX_NLOCAL)))
+#else
+#define MIGRAPHX_GLOBAL __global__
+#endif
 inline __device__ __attribute__((const)) index make_index()
 {
     return index{blockIdx.x * blockDim.x + threadIdx.x, threadIdx.x, blockIdx.x}; // NOLINT

src/targets/gpu/kernels/include/migraphx/kernels/reduce.hpp

@@ -174,6 +174,25 @@ struct inner_storage_tag
 template <class T>
 using is_inner_storage = is_base_of<inner_storage_tag, remove_cv_t<remove_reference_t<T>>>;
 
+template <class Size, class F>
+struct lazy_inner_storage : inner_storage_tag
+{
+    using type = remove_reference_t<decltype(declval<F>()(0, _c<0>))>;
+    F f;
+    constexpr Size rsize() const { return {}; }
+    template <class U, class V>
+    constexpr auto operator()(U j, V d) const
+    {
+        return f(j, d);
+    }
+};
+
+template <class Size, class F>
+constexpr lazy_inner_storage<Size, F> make_lazy_inner_storage(Size, F f)
+{
+    return {{}, f};
+}
+
 template <class R, class F>
 struct storage_access : F
 {
@@ -278,6 +297,14 @@ struct reducer_base
         });
     }
 
+    template <class F>
+    __device__ auto lazy_inner(F f) const
+    {
+        return this->inner_sliced([=](auto n, auto&&... xs) {
+            return make_lazy_inner_storage(n, [=](auto j, auto d) { return f(xs(j, d)...); });
+        });
+    }
+
     template <class Op, class T, class Read>
     __device__ auto reduce(Op op, T init, Read read) const
     {
@@ -396,25 +423,6 @@ struct block_large
         index idx;
         Slicer slice;
 
-        template <class Size, class F>
-        struct inner_storage : inner_storage_tag
-        {
-            using type = remove_reference_t<decltype(declval<F>()(0, _c<0>))>;
-            F f;
-            constexpr Size rsize() const { return {}; }
-            template <class U, class V>
-            constexpr auto operator()(U j, V d) const
-            {
-                return f(j, d);
-            }
-        };
-
-        template <class Size, class F>
-        static constexpr inner_storage<Size, F> make_inner_storage(Size, F f)
-        {
-            return {{}, {f}};
-        }
-
         template <class Op, class T, class Read, class N, class... Ts>
         __device__ auto reduce_impl(Op op, T init, Read read, N n, Ts&&... xs) const
         {
@@ -439,7 +447,7 @@ struct block_large
         template <class R, class F, class N, class... Ts>
         __device__ auto inner_impl(F f, N n, Ts&&... xs) const
         {
-            return make_inner_storage(n, [=](auto j, auto d) { return f(xs(j, d)...); });
+            return make_lazy_inner_storage(n, [=](auto j, auto d) { return f(xs(j, d)...); });
         }
     };
 
@@ -469,25 +477,6 @@ struct lane
         index idx;
         Slicer slice;
 
-        template <class Size, class F>
-        struct inner_storage : inner_storage_tag
-        {
-            using type = remove_reference_t<decltype(declval<F>()(0, _c<0>))>;
-            F f;
-            constexpr Size rsize() const { return {}; }
-            template <class U, class V>
-            constexpr auto operator()(U j, V d) const
-            {
-                return f(j, d);
-            }
-        };
-
-        template <class Size, class F>
-        static constexpr inner_storage<Size, F> make_inner_storage(Size, F f)
-        {
-            return {{}, {f}};
-        }
-
         template <class Op, class T, class Read, class N, class U, class... Us>
         __device__ auto reduce_impl(Op op, T init, Read read, N n, U&& x, Us&&... xs) const
         {
@@ -518,7 +507,7 @@ struct lane
         template <class R, class F, class N, class... Ts>
         __device__ auto inner_impl(F f, N n, Ts&&... xs) const
         {
-            return make_inner_storage(n, [=](auto j, auto d) { return f(xs(j, d)...); });
+            return make_lazy_inner_storage(n, [=](auto j, auto d) { return f(xs(j, d)...); });
         }
     };
     template <class Slicer>
@@ -577,5 +566,21 @@ simple_reduce(Op op, T init, Input input, Output output, ReadInput read, WriteOu
     });
 }
 
+template <class Algo, class Reduced, class Output, class F>
+__device__ void fused_reduce(Output output, F f)
+{
+    Algo::template run<Reduced>([&](auto out_idx, auto r) {
+        auto result = f(r);
+        if constexpr(reduce::is_inner_storage<decltype(result)>{})
+        {
+            r.inner([&](auto& y, auto x) { y = x; })(output, result);
+        }
+        else
+        {
+            r.outer([&] { output[out_idx] = implicit_conversion(result); });
+        }
+    });
+}
+
 } // namespace migraphx
 #endif // MIGRAPHX_GUARD_KERNELS_REDUCE_HPP

src/targets/gpu/mlir.cpp

@@ -197,10 +197,14 @@ struct mlir_program
                 result = mlirF64TypeGet(ctx.get());
             else if(as.is_integral())
             {
-                if(as.is_signed())
-                    result = mlirIntegerTypeSignedGet(ctx.get(), as.size() * 8);
-                else
-                    result = mlirIntegerTypeGet(ctx.get(), as.size() * 8);
+                // Note: rocMLIR use signless integer type for tensors types. This
+                // will translate to signed implementation for current supported
+                // operations.
+                if(as.is_unsigned())
+                {
+                    MIGRAPHX_THROW("Unsupported type: " + std::to_string(as.type_enum()));
+                }
+                result = mlirIntegerTypeGet(ctx.get(), as.size() * 8);
             }
             else
                 MIGRAPHX_THROW("Unsupported type: " + std::to_string(as.type_enum()));
@@ -483,7 +487,7 @@ struct mlir_program
     static value get_operator_value(const operation& op)
     {
         auto v = op.to_value();
-        if(op.name() == "convolution")
+        if(op.name() == "convolution" or op.name() == "quant_convolution")
         {
             // Adjust symetrical padding
             if(v.at("padding").size() == v.at("stride").size())

src/targets/gpu/target.cpp

@@ -32,6 +32,7 @@
 #include <migraphx/eliminate_identity.hpp>
 #include <migraphx/eliminate_pad.hpp>
 #include <migraphx/fuse_pointwise.hpp>
+#include <migraphx/fuse_reduce.hpp>
 #include <migraphx/inline_module.hpp>
 #include <migraphx/insert_pad.hpp>
 #include <migraphx/layout_nhwc.hpp>
@@ -72,6 +73,7 @@ namespace gpu {
 
 MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_DISABLE_SCHEDULE_PASS)
 MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_DISABLE_POINTWISE_FUSION)
+MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_DISABLE_REDUCE_FUSION)
 MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_ENABLE_NHWC)
 struct id_pass
 {
@@ -129,6 +131,8 @@ std::vector<pass> target::get_passes(migraphx::context& gctx, const compile_opti
         optimize_module{},
         enable_pass(not enabled(MIGRAPHX_DISABLE_POINTWISE_FUSION{}), fuse_pointwise{}),
         dead_code_elimination{},
+        enable_pass(not enabled(MIGRAPHX_DISABLE_REDUCE_FUSION{}), fuse_reduce{}),
+        dead_code_elimination{},
         fuse_mlir{&ctx},
         dead_code_elimination{},
         lowering{&ctx, options.offload_copy},