Merge branch 'develop' of github.com:ROCmSoftwarePlatform/AMDMIGraphX into dyn_shape_update

src/targets/gpu/lowering.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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 <iterator>
 #include <migraphx/gpu/lowering.hpp>
 #include <migraphx/manage_ptr.hpp>
@@ -58,7 +81,6 @@ struct miopen_apply
     const lowering* pass = nullptr;
     std::unordered_map<std::string, std::function<instruction_ref(instruction_ref)>> apply_map{};
     instruction_ref last{};
-    std::unordered_map<instruction_ref, std::string> prog_output_names{};
     bool offload_copy   = false;
     bool int8_x4_format = true;
     bool compute_fp32   = false;
@@ -77,27 +99,6 @@ struct miopen_apply
         (void)i;
     }
 
-    void create_output_names()
-    {
-        this->last = instruction::get_output_alias(std::prev(mod->end()));
-        if(this->last->name() == "@return")
-        {
-            const auto& prog_outputs = last->inputs();
-            std::vector<instruction_ref> outputs_alias(prog_outputs.size());
-
-            std::transform(prog_outputs.begin(),
-                           prog_outputs.end(),
-                           outputs_alias.begin(),
-                           [](const auto& i) { return instruction::get_output_alias(i); });
-
-            std::size_t index = 0;
-            for(auto ins : outputs_alias)
-            {
-                prog_output_names[ins] = mod->name() + ":#output_" + std::to_string(index++);
-            }
-        }
-    }
-
     const std::unordered_set<std::string>& get_rocblas_fp32_archs()
     {
         static std::unordered_set<std::string> supported_archs{"gfx908", "gfx90a"};
@@ -120,7 +121,6 @@ struct miopen_apply
 #endif
 
         offload_copy = (mod->name() == "main") ? pass->offload_copy : false;
-        create_output_names();
 
         add_generic_op("acos");
         add_generic_op("acosh");
@@ -201,7 +201,7 @@ struct miopen_apply
         add_quant_convolution_op();
     }
 
-    void copy_params()
+    void copy_params() const
     {
         if(not offload_copy)
             return;
@@ -261,7 +261,7 @@ struct miopen_apply
         copy_params();
     }
 
-    instruction_ref insert_precompile_op(instruction_ref ins)
+    instruction_ref insert_precompile_op(instruction_ref ins) const
     {
         auto output                       = insert_allocation(ins, ins->get_shape());
         std::vector<instruction_ref> refs = ins->inputs();
@@ -274,28 +274,9 @@ struct miopen_apply
             ins->module_inputs());
     }
 
-    instruction_ref insert_allocation(instruction_ref ins, const shape& s, std::string tag = "")
+    instruction_ref insert_allocation(instruction_ref ins, const shape& s) const
     {
-        // Instruction's output is an input of the ret instruction
-        if(offload_copy)
-        {
-            auto result = mod->insert_instruction(
-                ins, make_op("hip::allocate", {{"shape", to_value(s)}, {"tag", std::move(tag)}}));
-            return result;
-        }
-
-        auto ins_alias = instruction::get_output_alias(ins);
-        if(last->name() == "@return" and tag.empty() and prog_output_names.count(ins_alias) > 0)
-        {
-            return mod->add_parameter(prog_output_names[ins_alias], s);
-        }
-        else if(ins == last and tag.empty())
-        {
-            return mod->add_parameter("output", s);
-        }
-
-        return mod->insert_instruction(
-            ins, make_op("hip::allocate", {{"shape", to_value(s)}, {"tag", std::move(tag)}}));
+        return mod->insert_instruction(ins, make_op("allocate", {{"shape", to_value(s)}}));
     }
 
     void add_convolution_op()
@@ -306,7 +287,7 @@ struct miopen_apply
             auto conv = miopen_convolution{op, make_conv(op)};
             auto ws   = conv.find(get_context(), ins->get_shape(), to_shapes(ins->inputs()));
 
-            auto workspace = insert_allocation(ins, ws, "workspace");
+            auto workspace = insert_allocation(ins, ws);
             auto output    = insert_allocation(ins, ins->get_shape());
 
             return mod->replace_instruction(
@@ -322,7 +303,7 @@ struct miopen_apply
             auto conv = miopen_deconvolution{op, make_deconv(op)};
             auto ws   = conv.compile(get_context(), ins->get_shape(), to_shapes(ins->inputs()));
 
-            auto workspace = insert_allocation(ins, ws, "workspace");
+            auto workspace = insert_allocation(ins, ws);
             auto output    = insert_allocation(ins, ins->get_shape());
 
             return mod->replace_instruction(
@@ -335,27 +316,9 @@ struct miopen_apply
     {
         apply_map.emplace(name, [=](instruction_ref ins) {
             std::vector<instruction_ref> refs = ins->inputs();
-            if(refs.size() == 2)
-            {
-                auto output = insert_allocation(ins, ins->get_shape());
-                refs.push_back(output);
-            }
-            else
-            {
-                auto c_alias = instruction::get_output_alias(refs.back());
-                if(ins == last or refs.back()->outputs().size() > 1 or c_alias->inputs().empty())
-                {
-                    auto output = insert_allocation(ins, ins->get_shape());
-                    auto copy_out =
-                        mod->insert_instruction(ins, make_op("hip::copy"), refs.back(), output);
-                    refs.back() = copy_out;
-                    refs.push_back(copy_out);
-                }
-                else
-                {
-                    refs.push_back(refs.back());
-                }
-            }
+            assert(refs.size() == 2);
+            auto output = insert_allocation(ins, ins->get_shape());
+            refs.push_back(output);
             return mod->replace_instruction(
                 ins, rocblas_gemm<Op>{Op{}, 1, 0, int8_x4_format, compute_fp32}, refs);
         });
@@ -365,11 +328,25 @@ struct miopen_apply
     {
         apply_map.emplace("quant_convolution", [=](instruction_ref ins) {
             auto&& op = any_cast<op::quant_convolution>(ins->get_operator());
-            auto conv = miopen_quant_convolution{op, make_conv(op)};
-            auto ws   = conv.compile(get_context(), ins->get_shape(), to_shapes(ins->inputs()));
+            shape ws;
+            miopen_quant_convolution conv;
+            auto compile_quant_conv_with_format = [&](bool format) {
+                conv = miopen_quant_convolution{op, format, make_conv(op)};
+                ws   = conv.compile(get_context(), ins->get_shape(), to_shapes(ins->inputs()));
+            };
+
+            try
+            {
+                compile_quant_conv_with_format(int8_x4_format);
+            }
+            catch(migraphx::exception&)
+            {
+                // In case no solver supports the default format, retry using the other format.
+                compile_quant_conv_with_format(!int8_x4_format);
+            }
 
             auto args      = ins->inputs();
-            auto workspace = insert_allocation(ins, ws, "workspace");
+            auto workspace = insert_allocation(ins, ws);
             auto output    = insert_allocation(ins, ins->get_shape());
 
             return mod->replace_instruction(ins, conv, args[0], args[1], workspace, output);
@@ -466,33 +443,7 @@ struct miopen_apply
             auto sync_cond = mod->insert_instruction(ins, make_op("hip::sync_stream"), cpu_cond);
             inputs.front() = sync_cond;
 
-            std::vector<module_ref> mod_args = ins->module_inputs();
-            std::map<std::string, shape> name_shapes;
-            for(const auto& smod : mod_args)
-            {
-                auto ps = smod->get_parameter_shapes();
-                name_shapes.insert(ps.begin(), ps.end());
-            }
-
-            bool ins_output_allocated = false;
-            for(auto& pn : name_shapes)
-            {
-                const auto& s = pn.second;
-                instruction_ref output{};
-                if(s == ins->get_shape() and not ins_output_allocated)
-                {
-                    output               = insert_allocation(ins, s);
-                    ins_output_allocated = true;
-                }
-                else
-                {
-                    output = mod->insert_instruction(
-                        ins, make_op("hip::allocate", {{"shape", to_value(s)}}));
-                }
-                inputs.push_back(output);
-            }
-
-            return mod->replace_instruction(ins, ins->get_operator(), inputs, mod_args);
+            return mod->replace_instruction(ins, ins->get_operator(), inputs, ins->module_inputs());
         });
     }
 
@@ -511,20 +462,17 @@ struct miopen_apply
             inputs.at(0)     = synced_max_iter;
             inputs.at(1)     = cpu_cond;
             auto copy_inputs = inputs;
-            std::transform(
-                copy_inputs.begin(), copy_inputs.end(), std::back_inserter(inputs), [&](auto in) {
-                    return mod->insert_instruction(
-                        ins, make_op("hip::allocate", {{"shape", to_value(in->get_shape())}}));
-                });
+            std::transform(copy_inputs.begin(),
+                           copy_inputs.end(),
+                           std::back_inserter(inputs),
+                           [&](auto in) { return insert_allocation(ins, in->get_shape()); });
 
             auto mod_args = ins->module_inputs();
             auto output   = insert_allocation(ins, ins->get_shape());
 
             const auto* sub_mod = mod_args.front();
-            auto cond_out       = mod->insert_instruction(
-                ins,
-                make_op("hip::allocate",
-                        {{"shape", to_value(sub_mod->get_output_shapes().front())}}));
+            auto cond_out       = insert_allocation(ins, sub_mod->get_output_shapes().front());
+
             // add cond and mod outputs to the argument list
             inputs.push_back(cond_out);
             inputs.push_back(output);

src/targets/gpu/lrn.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/lrn.hpp>
 #include <migraphx/gpu/context.hpp>
 

src/targets/gpu/mlir_conv.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/mlir_conv.hpp>
 #include <migraphx/manage_ptr.hpp>
 #include <migraphx/instruction.hpp>

src/targets/gpu/multinomial.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/multinomial.hpp>
 #include <migraphx/gpu/device/multinomial.hpp>
 #include <migraphx/gpu/context.hpp>

src/targets/gpu/nonzero.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/nonzero.hpp>
 #include <migraphx/gpu/context.hpp>
 #include <migraphx/gpu/device/nonzero.hpp>

src/targets/gpu/pack_args.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/pack_args.hpp>
 #include <migraphx/requires.hpp>
 

src/targets/gpu/pack_int8_args.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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 <iterator>
 #include <migraphx/gpu/pack_int8_args.hpp>
 #include <migraphx/gpu/int8_gemm_pack.hpp>
@@ -22,10 +45,10 @@ static instruction_ref pad_ins(module& m, instruction_ref ins, int offset)
     auto pad_k                     = (k + 3) / 4 * 4;
     auto pad_lens                  = lens;
     pad_lens[lens.size() + offset] = pad_k;
-    std::vector<int64_t> pad_dims(lens.size() * 2, 0);
-    auto ret_ins = ins;
+    auto ret_ins                   = ins;
     if(pad_k != k)
     {
+        std::vector<int64_t> pad_dims(lens.size() * 2, 0);
         pad_dims[lens.size() + offset] = pad_k - k;
         shape ps{s.type(), pad_lens};
         auto ins_out =
@@ -118,7 +141,7 @@ void pack_int8_args::apply(module& m) const
             assert(val.contains("int8_x4_format"));
             if(not val.at("int8_x4_format").to<bool>())
             {
-                return;
+                continue;
             }
             auto inputs = ins->inputs();
             auto lens   = inputs.at(0)->get_shape().lens();
@@ -156,6 +179,12 @@ void pack_int8_args::apply(module& m) const
         }
         else if(ins->name() == "gpu::quant_convolution")
         {
+            auto val = ins->get_operator().to_value();
+            if(not val.at("int8_x4_format").to<bool>())
+            {
+                continue;
+            }
+
             auto inputs   = ins->inputs();
             auto packed_x = m.insert_instruction(
                 ins,

src/targets/gpu/pad.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/pad.hpp>
 #include <migraphx/gpu/context.hpp>
 #include <migraphx/gpu/device/pad.hpp>

src/targets/gpu/pooling.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/pooling.hpp>
 #include <migraphx/gpu/context.hpp>
 

src/targets/gpu/prefuse_ops.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/prefuse_ops.hpp>
 #include <migraphx/match/layernorm.hpp>
 #include <migraphx/make_op.hpp>

src/targets/gpu/quant_convolution.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/quant_convolution.hpp>
 #include <migraphx/gpu/device/convert.hpp>
 #include <migraphx/gpu/context.hpp>
@@ -16,8 +39,8 @@ argument miopen_quant_convolution::compute(context& ctx,
                                            const shape& output_shape,
                                            const std::vector<argument>& args) const
 {
-    auto x_desc = make_tensor(args[0].get_shape(), true);
-    auto w_desc = make_tensor(args[1].get_shape(), true);
+    auto x_desc = make_tensor(args[0].get_shape(), int8_x4_format);
+    auto w_desc = make_tensor(args[1].get_shape(), int8_x4_format);
     auto y_desc = make_tensor(output_shape);
 
     float alpha = 1;
@@ -49,8 +72,8 @@ shape miopen_quant_convolution::compile(context& ctx,
                                         std::vector<shape> inputs)
 {
     shape workspace_shape{};
-    auto x_desc = make_tensor(inputs[0], true);
-    auto w_desc = make_tensor(inputs[1], true);
+    auto x_desc = make_tensor(inputs[0], int8_x4_format);
+    auto w_desc = make_tensor(inputs[1], int8_x4_format);
     auto y_desc = make_tensor(output_shape);
 
     std::size_t workspace_size = 0;
@@ -62,8 +85,15 @@ shape miopen_quant_convolution::compile(context& ctx,
                                              &workspace_size);
     workspace_shape = shape{shape::int8_type, {workspace_size}};
 
-    auto arg_vec4_x = to_gpu(generate_argument(pack_int8_shape(inputs[0])));
-    auto arg_vec4_w = to_gpu(generate_argument(pack_int8_shape(inputs[1])));
+    auto x_shape = inputs[0];
+    auto w_shape = inputs[1];
+    if(int8_x4_format)
+    {
+        x_shape = pack_int8_shape(x_shape);
+        w_shape = pack_int8_shape(w_shape);
+    }
+    auto arg_vec4_x = to_gpu(generate_argument(x_shape));
+    auto arg_vec4_w = to_gpu(generate_argument(w_shape));
     auto y          = allocate_gpu(output_shape);
     auto workspace  = allocate_gpu(workspace_shape);
 

src/targets/gpu/reverse.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/reverse.hpp>
 #include <migraphx/gpu/context.hpp>
 #include <migraphx/gpu/device/reverse.hpp>

src/targets/gpu/rnn_variable_seq_lens.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/rnn_variable_seq_lens.hpp>
 #include <migraphx/gpu/context.hpp>
 #include <migraphx/gpu/device/rnn_variable_seq_lens.hpp>

src/targets/gpu/rocblas.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/rocblas.hpp>
 
 namespace migraphx {

src/targets/gpu/scatter.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/scatter.hpp>
 #include <migraphx/gpu/context.hpp>
 #include <migraphx/gpu/device/scatter.hpp>

src/targets/gpu/schedule_model.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/schedule_model.hpp>
 #include <migraphx/gpu/context.hpp>
 #include <migraphx/register_op.hpp>

src/targets/gpu/softmax.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/softmax.hpp>
 #include <migraphx/gpu/device/softmax.hpp>
 #include <migraphx/gpu/context.hpp>

src/targets/gpu/sync_device.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/sync_device.hpp>
 #include <migraphx/program.hpp>
 #include <migraphx/instruction.hpp>

src/targets/gpu/target.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/adjust_allocation.hpp>
 #include <migraphx/auto_contiguous.hpp>
 #include <migraphx/check_context.hpp>
@@ -17,6 +40,7 @@
 #include <migraphx/preallocate_param.hpp>
 #include <migraphx/propagate_constant.hpp>
 #include <migraphx/register_target.hpp>
+#include <migraphx/replace_allocate.hpp>
 #include <migraphx/rewrite_batchnorm.hpp>
 #include <migraphx/rewrite_pooling.hpp>
 #include <migraphx/rewrite_quantization.hpp>
@@ -29,7 +53,6 @@
 #include <migraphx/gpu/compile_ops.hpp>
 #include <migraphx/gpu/concat_gpu_opt.hpp>
 #include <migraphx/gpu/context.hpp>
-#include <migraphx/gpu/eliminate_workspace.hpp>
 #include <migraphx/gpu/fuse_ops.hpp>
 #include <migraphx/gpu/prefuse_ops.hpp>
 #include <migraphx/gpu/lowering.hpp>
@@ -109,6 +132,8 @@ std::vector<pass> target::get_passes(migraphx::context& gctx, const compile_opti
         lowering{&ctx, options.offload_copy},
         eliminate_contiguous{"gpu::contiguous"},
         dead_code_elimination{},
+        replace_allocate{gpu_allocation_model{}, options.offload_copy},
+        dead_code_elimination{},
         eliminate_concat{concat_gpu_optimization{}},
         dead_code_elimination{},
         pack_int8_args{},
@@ -125,7 +150,6 @@ std::vector<pass> target::get_passes(migraphx::context& gctx, const compile_opti
         sync_device{},
         preallocate_param{"scratch", gpu_allocation_model{}},
         dead_code_elimination{},
-        eliminate_workspace{},
         eliminate_allocation{"hip::allocate"},
         check_context<context>{},
         normalize_ops{},

src/targets/gpu/topk.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 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/topk.hpp>
 #include <migraphx/gpu/context.hpp>
 #include <migraphx/gpu/device/topk.hpp>