Support leaky_relu in mlir

src/targets/cpu/lowering.cpp

@@ -340,7 +340,7 @@ struct cpu_apply
         extend_op("softmax", "dnnl::softmax");
 
         extend_op("im2col", "cpu::im2col", false);
-        extend_op("leaky_relu", "cpu::leaky_relu", false);
+        // extend_op("leaky_relu", "cpu::leaky_relu", false);
         extend_op("pad", "cpu::pad", false);
         extend_op("rnn_var_sl_last_output", "cpu::rnn_var_sl_last_output", false);
     }

src/targets/gpu/CMakeLists.txt

@@ -175,6 +175,7 @@ add_library(migraphx_gpu
     nonzero.cpp
     pack_args.cpp
     prefuse_ops.cpp
+    prepare_mlir.cpp
     prepare_reduce.cpp
     perfdb.cpp
     pooling.cpp

src/targets/gpu/fuse_mlir.cpp

@@ -463,22 +463,23 @@
     }
     const std::initializer_list<std::string> any_type_ops = {"@literal", "@param", "@return"};
     const std::initializer_list<std::string> no_bool_ops  = {
+        "abs",
+        "add",
+        "clip",
         "convolution",
-        "quant_convolution",
+        "dequantizelinear",
+        "div",
         "dot",
+        "leaky_relu",
+        "mul",
+        "neg",
+        "pow",
+        "quant_convolution",
         "quant_dot",
-        "add",
-        "clip",
+        "quantizelinear",
         "relu",
         "sub",
-        "mul",
-        "div",
-        "pow",
         "where",
-        "quantizelinear",
-        "dequantizelinear",
-        "abs",
-        "neg",
     };
     const std::initializer_list<std::string> fp_only_ops = {
         "ceil",
@@ -552,6 +553,8 @@
     return is_pointwise_op_supported_by_mlir(i);
 }
 
+static bool is_reduce(const instruction& ins) { return contains(ins.name(), "reduce"); }
+
 MIGRAPHX_PRED_MATCHER(mlir_split_reduce, instruction_ref ins)
 {
     if(ins->name() != "split_fused_reduce")

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

@@ -1,7 +1,7 @@
 /*
  * The MIT License (MIT)
  *
- * Copyright (c) 2015-2024 Advanced Micro Devices, Inc. All rights reserved.
+ * Copyright (c) 2015-2025 Advanced Micro Devices, Inc. All rights reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a copy
  * of this software and associated documentation files (the "Software"), to deal
@@ -53,8 +53,6 @@ struct MIGRAPHX_GPU_EXPORT mlir_code_object
     std::vector<value> prefill_values   = {};
 };
 
-MIGRAPHX_GPU_EXPORT bool is_reduce(const instruction& ins);
-
 MIGRAPHX_GPU_EXPORT mlir_code_object compile_mlir(const context& migraphx_ctx,
                                                   module m,
                                                   const std::vector<shape>& in_shapes,

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

@@ -0,0 +1,47 @@
+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2025 Advanced Micro Devices, Inc. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ */
+#ifndef MIGRAPHX_GUARD_GPU_PREPARE_MLIR_HPP
+#define MIGRAPHX_GUARD_GPU_PREPARE_MLIR_HPP
+
+#include <migraphx/config.hpp>
+#include <string>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+
+struct module;
+
+namespace gpu {
+
+struct prepare_mlir
+{
+    std::string name() const { return "gpu::prepare_mlir"; }
+    void apply(module& m) const;
+};
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+#endif // MIGRAPHX_GUARD_GPU_PREPARE_REDUCE_HPP

src/targets/gpu/mlir.cpp

@@ -31,6 +31,7 @@
 #include <migraphx/dead_code_elimination.hpp>
 #include <migraphx/pass_manager.hpp>
 #include <migraphx/gpu/mlir.hpp>
+#include <migraphx/gpu/prepare_mlir.hpp>
 #include <mlir-c/Dialect/RockEnums.h>
 #include <numeric>
 #include <ostream>
@@ -1070,58 +1071,12 @@ struct mlir_program
     std::string sym_name;
 };
 
-bool is_reduce(const instruction& ins) { return contains(ins.name(), "reduce"); }
-
-static void rewrite_reduce(module& m)
-{
-    for(auto i : iterator_for(m))
-    {
-        if(is_reduce(*i))
-        {
-            auto reduce_op      = i->get_operator().to_value();
-            auto reduce_op_name = i->get_operator().name();
-            auto reduce_axes    = reduce_op["axes"].to_vector<size_t>();
-            auto reduce_lens    = i->get_shape().lens();
-            auto in_shape       = i->inputs().front()->get_shape();
-            const auto& in_lens = in_shape.lens();
-            assert(in_shape.standard());
-            assert(reduce_lens.size() == in_lens.size());
-            assert(std::adjacent_find(
-                       reduce_axes.begin(), reduce_axes.end(), [](auto axis_1, auto axis_2) {
-                           return axis_2 - axis_1 > 1;
-                       }) == reduce_axes.end());
-
-            std::vector<int64_t> new_rsp_dims;
-            std::vector<int64_t> new_reduce_axes;
-            for(const auto axis : range(in_shape.ndim()))
-            {
-                if(reduce_lens[axis] == in_lens[axis])
-                {
-                    new_rsp_dims.push_back(in_lens[axis]);
-                }
-                else if(new_reduce_axes.empty())
-                {
-                    assert(reduce_lens[axis] == 1);
-                    new_rsp_dims.push_back(-1);
-                    new_reduce_axes.push_back(axis);
-                }
-            }
-            auto rsp_ins = m.insert_instruction(
-                i, migraphx::make_op("reshape", {{"dims", new_rsp_dims}}), i->inputs().front());
-            auto collapsed_reduce = m.insert_instruction(
-                i, migraphx::make_op(reduce_op_name, {{"axes", new_reduce_axes}}), rsp_ins);
-            auto rsp_back = m.insert_instruction(
-                i, migraphx::make_op("reshape", {{"dims", reduce_lens}}), collapsed_reduce);
-            m.replace_instruction(i, rsp_back);
-        }
-    }
-    migraphx::run_passes(m, {migraphx::dead_code_elimination{}});
-}
+static void prepare(module& m) { run_passes(m, {prepare_mlir{}}); }
 
 bool is_module_fusible(const module& m, const context& migraphx_ctx, const value& solution)
 {
     auto mm = m;
-    rewrite_reduce(mm);
+    prepare(mm);
     mlir_program mp;
     mp.set_gpu_properties(migraphx_ctx);
     mp.parse(mm);
@@ -1171,7 +1126,7 @@ std::string dump_mlir(module m, const std::vector<shape>& inputs)
     {
         adjust_param_shapes(m, inputs);
     }
-    rewrite_reduce(m);
+    prepare(m);
     mlir_program mp;
     mp.parse(*mr);
     auto mod_op = mlirModuleGetOperation(mp.mmodule.get());
@@ -1232,7 +1187,7 @@ void dump_mlir_to_file(module m, const std::vector<shape>& inputs, const fs::pat
     {
         adjust_param_shapes(m, inputs);
     }
-    rewrite_reduce(m);
+    prepare(m);
 
     auto name = compute_dump_name(m, ".mlir");
     auto f    = location / name;
@@ -1255,7 +1210,7 @@ mlir_code_object compile_mlir(const context& migraphx_ctx,
                               const value& solution)
 {
     adjust_param_shapes(m, in_shapes);
-    rewrite_reduce(m);
+    prepare(m);
     const bool trace = enabled(MIGRAPHX_TRACE_MLIR{});
 
     static std::mutex mutex;
@@ -1336,7 +1291,7 @@ tuning_config get_tuning_config_mlir(const context& migraphx_ctx,
                                      bool exhaustive)
 {
     adjust_param_shapes(m, inputs);
-    rewrite_reduce(m);
+    prepare(m);
     mlir_program mp;
     mp.set_gpu_properties(migraphx_ctx);
     mp.parse(m);

src/targets/gpu/prepare_mlir.cpp

@@ -0,0 +1,143 @@
+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2025 Advanced Micro Devices, Inc. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ */
+#include <migraphx/gpu/prepare_mlir.hpp>
+#include <migraphx/common.hpp>
+#include <migraphx/dead_code_elimination.hpp>
+#include <migraphx/instruction.hpp>
+#include <migraphx/iterator_for.hpp>
+#include <migraphx/make_op.hpp>
+#include <migraphx/matcher.hpp>
+#include <migraphx/module.hpp>
+#include <migraphx/pass_manager.hpp>
+#include <migraphx/ranges.hpp>
+#include <algorithm>
+#include <numeric>
+#include <vector>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+
+namespace {
+
+struct find_reduce
+{
+    auto matcher() const { return match::name_contains("reduce"); }
+
+    void apply(module& m, const match::matcher_result& r) const
+    {
+        auto ins            = r.result;
+        auto reduce_op      = ins->get_operator().to_value();
+        auto reduce_op_name = ins->get_operator().name();
+        auto reduce_axes    = reduce_op["axes"].to_vector<size_t>();
+        auto reduce_lens    = ins->get_shape().lens();
+        auto in_shape       = ins->inputs().front()->get_shape();
+        const auto& in_lens = in_shape.lens();
+        assert(in_shape.standard());
+        assert(reduce_lens.size() == in_lens.size());
+        assert(std::adjacent_find(
+                   reduce_axes.begin(), reduce_axes.end(), [](auto axis_1, auto axis_2) {
+                       return axis_2 - axis_1 > 1;
+                   }) == reduce_axes.end());
+
+        std::vector<int64_t> new_rsp_dims;
+        std::vector<int64_t> new_reduce_axes;
+        for(const auto axis : range(in_shape.ndim()))
+        {
+            if(reduce_lens[axis] == in_lens[axis])
+            {
+                new_rsp_dims.push_back(in_lens[axis]);
+            }
+            else if(new_reduce_axes.empty())
+            {
+                assert(reduce_lens[axis] == 1);
+                new_rsp_dims.push_back(-1);
+                new_reduce_axes.push_back(axis);
+            }
+        }
+        auto rsp_ins = m.insert_instruction(
+            ins, migraphx::make_op("reshape", {{"dims", new_rsp_dims}}), ins->inputs().front());
+        auto collapsed_reduce = m.insert_instruction(
+            ins, migraphx::make_op(reduce_op_name, {{"axes", new_reduce_axes}}), rsp_ins);
+        auto rsp_back = m.insert_instruction(
+            ins, migraphx::make_op("reshape", {{"dims", reduce_lens}}), collapsed_reduce);
+        m.replace_instruction(ins, rsp_back);
+    }
+};
+
+struct find_leaky_relu
+{
+    auto matcher() const { return match::name("leaky_relu"); }
+
+    void apply(module& m, const match::matcher_result& r) const
+    {
+        auto ins   = r.result;
+        auto x_ins = ins->inputs().front();
+
+        float alpha_f = ins->get_operator().to_value()["alpha"].to<float>();
+        auto alpha    = m.add_literal(literal{{x_ins->get_shape().type(), {1}}, {alpha_f}});
+        auto zero     = m.add_literal(literal{{x_ins->get_shape().type(), {1}}, {0.0}});
+
+        auto greater   = insert_common_op(m, ins, make_op("greater"), {x_ins, zero});
+        auto mul_alpha = insert_common_op(m, ins, make_op("mul"), {x_ins, alpha});
+
+        m.replace_instruction(ins, make_op("where"), {greater, x_ins, mul_alpha});
+    }
+};
+
+// mlir has issues sometime when the condition to `where` is not a bool. So this will convert the
+// condition to a bool.
+struct find_where
+{
+    auto matcher() const { return match::name("where"); }
+
+    void apply(module& m, const match::matcher_result& r) const
+    {
+        auto ins      = r.result;
+        auto cond_ins = ins->inputs().front();
+
+        if(cond_ins->get_shape().type() == shape::bool_type)
+            return;
+
+        auto bool_cond_ins = m.insert_instruction(
+            ins, make_op("convert", {{"target_type", shape::bool_type}}), cond_ins);
+
+        m.replace_instruction(
+            ins, make_op("where"), {bool_cond_ins, ins->inputs()[1], ins->inputs()[2]});
+    }
+};
+
+} // namespace
+
+void prepare_mlir::apply(module& m) const
+{
+    match::find_matches(m, find_reduce{}, find_leaky_relu{});
+    match::find_matches(m, find_where{});
+    run_passes(m, {dead_code_elimination{}});
+}
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx