Merge branch 'develop' into resnet50_partition

Jenkinsfile

@@ -107,7 +107,7 @@ def rocmnode(name, body) {
     }
 }
 
-rocmtest clang_debug: rocmnode('cdna') { cmake_build ->
+rocmtest clang_debug: rocmnode('mi100+') { cmake_build ->
     stage('hipRTC Debug') {
         def sanitizers = "undefined"
         def debug_flags = "-g -O2 -fsanitize=${sanitizers} -fno-sanitize-recover=${sanitizers}"
@@ -124,12 +124,12 @@ rocmtest clang_debug: rocmnode('cdna') { cmake_build ->
 //     stage('Hidden symbols') {
 //         cmake_build(flags: "-DMIGRAPHX_ENABLE_PYTHON=Off -DMIGRAPHX_ENABLE_GPU=On -DMIGRAPHX_ENABLE_CPU=On -DCMAKE_CXX_VISIBILITY_PRESET=hidden -DCMAKE_C_VISIBILITY_PRESET=hidden")
 //     }
-}, all_targets_debug : rocmnode('cdna') { cmake_build ->
+}, all_targets_debug : rocmnode('mi100+') { cmake_build ->
     stage('All targets Release') {
         def gpu_targets = getgputargets()
         cmake_build(flags: "-DCMAKE_BUILD_TYPE=release -DMIGRAPHX_ENABLE_GPU=On -DMIGRAPHX_ENABLE_CPU=On -DMIGRAPHX_ENABLE_FPGA=On -DGPU_TARGETS='${gpu_targets}'")
     }
-}, mlir_debug: rocmnode('cdna') { cmake_build ->
+}, mlir_debug: rocmnode('mi100+') { cmake_build ->
     stage('MLIR Debug') {
         withEnv(['MIGRAPHX_ENABLE_MLIR=1']) {
             def sanitizers = "undefined"

src/driver/main.cpp

@@ -540,17 +540,20 @@ struct params : command<params>
 struct verify : command<verify>
 {
     compiler c;
-    migraphx::verify::tolerance tols;
+    // Set to -1. as nonsense initial value
+    double rms_tol       = -1.0;
+    double atol          = -1.0;
+    double rtol          = -1.0;
     bool per_instruction = false;
     bool reduce          = false;
     void parse(argument_parser& ap)
     {
         c.parse(ap);
-        ap(tols.rms_tol, {"--rms-tol"}, ap.help("Tolerance for the RMS error (Default: 0.001)"));
-        ap(tols.atol,
+        ap(rms_tol, {"--rms-tol"}, ap.help("Tolerance for the RMS error (Default: 0.001)"));
+        ap(atol,
            {"--atol"},
            ap.help("Tolerance for the elementwise absolute difference (Default: 0.001)"));
-        ap(tols.rtol,
+        ap(rtol,
            {"--rtol"},
            ap.help("Tolerance for the elementwise relative difference (Default: 0.001)"));
         ap(per_instruction,
@@ -569,11 +572,42 @@ struct verify : command<verify>
         auto t = c.ct.get_target();
         auto m = c.parameters.generate(p, t, true, c.l.batch);
 
+        // TODO remove this and make the driver able to figure out datatype most used in the model
+        //  then set the tolerances appropriately. Need to check here because c.to_fp16 only set
+        //  after argument_parser.parse() is run. This code is complicated because there's not a
+        //  good way to change the default tolerances after reading `--fp16` but before reading
+        //  `--rms-tol`, `--atol`, and `--rtol`.
+        migraphx::verify::tolerance tols{};
+        if(c.to_fp16)
+        {
+            tols = migraphx::verify::tolerance{8e-2, 4e-2, 4e-2};
+        }
+        if(not float_equal(this->rms_tol, -1.0))
+        {
+            tols.rms_tol = this->rms_tol;
+        }
+        if(not float_equal(this->atol, -1.0))
+        {
+            tols.atol = this->atol;
+        }
+        if(not float_equal(this->rtol, -1.0))
+        {
+            tols.rtol = this->rtol;
+        }
+
+        std::cout << "rms_tol: " << tols.rms_tol << std::endl;
+        std::cout << "atol: " << tols.atol << std::endl;
+        std::cout << "rtol: " << tols.rtol << std::endl;
+
         auto quantize = precision::fp32;
         if(c.to_fp16)
+        {
             quantize = precision::fp16;
+        }
         if(c.to_int8)
+        {
             quantize = precision::int8;
+        }
 
         if(per_instruction)
         {

src/targets/gpu/fuse_mlir.cpp

@@ -131,9 +131,16 @@ fuse_input_ops_and_gemm_based_op(module_ref mm, instruction_ref gemm_based_op)
     for(instruction_ref input : gemm_based_op->inputs())
     {
         std::vector<operation> op_stream;
-        while(contains({"slice", "transpose", "contiguous", "reshape"}, input->name()))
+        while(contains(
+            {"slice", "transpose", "contiguous", "reshape", "squeeze", "flatten", "unsqueeze"},
+            input->name()))
         {
-            op_stream.push_back(input->get_operator());
+            operation op = input->get_operator();
+            if(contains({"squeeze", "flatten", "unsqueeze"}, input->name()))
+            {
+                op = migraphx::make_op("reshape", {{"dims", input->get_shape().lens()}});
+            }
+            op_stream.push_back(op);
             input = input->inputs().at(0);
         }
         top_inputs.push_back(input);

test/py/test_gpu.py

 #####################################################################################
 # The MIT License (MIT)
 #
-# Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
+# Copyright (c) 2015-2023 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
@@ -21,12 +21,8 @@
 # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 # THE SOFTWARE.
 #####################################################################################
-import sys
 import migraphx
-try:
-    import numpy as np
-except:
-    sys.exit()
+import numpy as np
 
 
 def test_conv_relu():

test/py/test_numpy.py

 #####################################################################################
 # The MIT License (MIT)
 #
-# Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
+# Copyright (c) 2015-2023 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
@@ -21,11 +21,8 @@
 # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 # THE SOFTWARE.
 #####################################################################################
-import migraphx, sys
-try:
-    import numpy as np
-except:
-    sys.exit()
+import migraphx
+import numpy as np
 
 
 def test_add_op():

test/verify/test_flatten_dot_relu.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 "verify_program.hpp"
+#include <migraphx/program.hpp>
+#include <migraphx/generate.hpp>
+#include <migraphx/make_op.hpp>
+
+struct test_flatten_dot_relu : verify_program<test_flatten_dot_relu>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        auto* mm = p.get_main_module();
+        auto a =
+            mm->add_parameter("a", migraphx::shape{migraphx::shape::float_type, {1, 2, 3, 3, 5}});
+        a = mm->add_instruction(migraphx::make_op("flatten", {{"axis", 3}}), a);
+        auto b =
+            mm->add_parameter("b", migraphx::shape{migraphx::shape::float_type, {1, 5, 3, 3, 1}});
+        b        = mm->add_instruction(migraphx::make_op("flatten", {{"axis", 3}}), b);
+        auto dot = mm->add_instruction(migraphx::make_op("dot"), a, b);
+        mm->add_instruction(migraphx::make_op("relu"), dot);
+        return p;
+    }
+};

test/verify/test_squeeze_conv_relu.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 "verify_program.hpp"
+#include <migraphx/program.hpp>
+#include <migraphx/generate.hpp>
+#include <migraphx/make_op.hpp>
+
+struct test_squeeze_conv_relu : verify_program<test_squeeze_conv_relu>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        auto* mm = p.get_main_module();
+        auto input =
+            mm->add_parameter("x", migraphx::shape{migraphx::shape::float_type, {4, 3, 1, 3, 3}});
+        input = mm->add_instruction(migraphx::make_op("squeeze", {{"axes", {2}}}), input);
+        auto weights =
+            mm->add_parameter("w", migraphx::shape{migraphx::shape::float_type, {4, 3, 3, 3}});
+        auto conv = mm->add_instruction(migraphx::make_op("convolution"), input, weights);
+        mm->add_instruction(migraphx::make_op("relu"), conv);
+        return p;
+    }
+};

test/verify/test_unsqueeze_conv_relu.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 "verify_program.hpp"
+#include <migraphx/program.hpp>
+#include <migraphx/generate.hpp>
+#include <migraphx/make_op.hpp>
+
+struct test_unsqueeze_conv_relu : verify_program<test_unsqueeze_conv_relu>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        auto* mm = p.get_main_module();
+        auto input =
+            mm->add_parameter("x", migraphx::shape{migraphx::shape::float_type, {1, 3, 3, 3}});
+        auto weights =
+            mm->add_parameter("w", migraphx::shape{migraphx::shape::float_type, {3, 3, 3}});
+        weights   = mm->add_instruction(migraphx::make_op("unsqueeze", {{"axes", {0}}}), weights);
+        auto conv = mm->add_instruction(migraphx::make_op("convolution"), input, weights);
+        mm->add_instruction(migraphx::make_op("relu"), conv);
+        return p;
+    }
+};

tools/CMakeLists.txt

@@ -28,7 +28,7 @@ if(NOT Python_EXECUTABLE)
     return()
 endif()
 
-find_program(CLANG_FORMAT clang-format PATHS /opt/rocm/llvm $ENV{HIP_PATH})
+find_program(CLANG_FORMAT clang-format PATHS /opt/rocm/llvm ENV HIP_PATH PATH_SUFFIXES bin)
 if(NOT CLANG_FORMAT)
     message(WARNING "clang-format not found - skipping 'generate' target!")
     return()