Adding stress test for autogeneration and autocorrection, and fixing preshuffle bug.

Author: vidyasagar-amd

dispatcher/examples/conv/cpp/02_conv_validation.cpp

@@ -2,13 +2,13 @@
 // Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
 
 /**
- * Example 03: Convolution with CPU Validation - Declarative
+ * Example 02: Convolution with CPU Validation - Declarative
  *
  * Demonstrates convolution with CPU reference verification.
  * Uses the Signature/Algorithm/Arch declarative pattern.
  *
  * Self-contained build:
- *   python3 scripts/compile_conv_examples.py examples/conv/cpp/03_conv_validation.cpp
+ *   python3 scripts/compile_conv_examples.py examples/conv/cpp/02_conv_validation.cpp
  *
  * Complexity: ★★★☆☆
  */
@@ -22,6 +22,7 @@
 
 // Declarative utilities
 #include "ck_tile/dispatcher/conv_utils.hpp"
+#include "ck_tile/dispatcher/example_args.hpp"
 
 // CK Tile includes
 #include "ck_tile/core.hpp"
@@ -32,6 +33,7 @@
 
 using namespace ck_tile::dispatcher;
 using namespace ck_tile::dispatcher::conv_utils;
+using namespace ck_tile::dispatcher::utils;
 
 // =============================================================================
 // KERNEL DECLARATIONS
@@ -63,10 +65,28 @@ using AccDataType = float;
 
 int main(int argc, char* argv[])
 {
+    ExampleArgs args("Example 02: Conv Validation", "Convolution with CPU reference verification");
+    args.add_option("-n", "1", "Batch size N");
+    args.add_option("-c", "64", "Input channels C");
+    args.add_option("-k", "128", "Output channels K");
+    args.add_option("--size", "14", "Spatial size (H=W)");
+    args.add_flag("--no-verify", "Skip CPU validation");
+    args.add_flag("--list", "List all kernel sets");
+
+    if(!args.parse(argc, argv))
+        return 0;
+
     std::cout << "======================================================================\n";
-    std::cout << "Example 03: Convolution with CPU Validation (Declarative)\n";
+    std::cout << "Example 02: Convolution with CPU Validation (Declarative)\n";
     std::cout << "======================================================================\n\n";
 
+    if(args.has("--list"))
+    {
+        std::cout << "Declared Kernel Sets:\n";
+        ConvKernelSetRegistry::instance().print();
+        return 0;
+    }
+
     // -------------------------------------------------------------------------
     // Step 1: Show declared kernels
     // -------------------------------------------------------------------------
@@ -83,23 +103,13 @@ int main(int argc, char* argv[])
     std::cout << "Step 2: Define Problem\n";
     std::cout << "----------------------\n";
 
-    int N = 1, C = 64, K = 128, Hi = 14, Wi = 14, Y = 3, X = 3;
-    bool verify = true;
-
-    for(int i = 1; i < argc; ++i)
-    {
-        std::string arg = argv[i];
-        if(arg == "-n" && i + 1 < argc)
-            N = std::stoi(argv[++i]);
-        else if(arg == "-c" && i + 1 < argc)
-            C = std::stoi(argv[++i]);
-        else if(arg == "-k" && i + 1 < argc)
-            K = std::stoi(argv[++i]);
-        else if(arg == "-h" && i + 1 < argc)
-            Hi = Wi = std::stoi(argv[++i]);
-        else if(arg == "--no-verify")
-            verify = false;
-    }
+    int N  = args.get_int("-n", 1);
+    int C  = args.get_int("-c", 64);
+    int K  = args.get_int("-k", 128);
+    int Hi = args.get_int("--size", 14);
+    int Wi = Hi;
+    int Y = 3, X = 3;
+    bool verify = !args.has("--no-verify");
 
     auto problem = create_conv2d_problem(N, C, K, Hi, Wi, Y, X, 1, 1, ConvOp::Forward);
     print_problem(problem);

dispatcher/examples/conv/cpp/03_multi_size.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
 
 /**
- * Example 04: Multi-Size Convolution with GPU Execution
+ * Example 03: Multi-Size Convolution with GPU Execution
  *
  * Demonstrates using different kernel tile sizes for different problem sizes,
  * with actual GPU execution for each.
@@ -16,13 +16,15 @@
 #include <hip/hip_runtime.h>
 
 #include "ck_tile/dispatcher/conv_utils.hpp"
+#include "ck_tile/dispatcher/example_args.hpp"
 #include "ck_tile/core.hpp"
 #include "ck_tile/host.hpp"
 #include "ck_tile/host/convolution_parameter.hpp"
 #include "ck_tile/ops/grouped_convolution.hpp"
 
 using namespace ck_tile::dispatcher;
 using namespace ck_tile::dispatcher::conv_utils;
+using namespace ck_tile::dispatcher::utils;
 
 // =============================================================================
 // KERNEL DECLARATIONS - Multiple tile sizes
@@ -131,12 +133,26 @@ void run_conv_on_gpu(const ConvProblem& problem, const std::string& label)
 // MAIN
 // =============================================================================
 
-int main()
+int main(int argc, char* argv[])
 {
+    ExampleArgs args("Example 03: Multi-Size Conv",
+                     "Different tile sizes for different problem sizes");
+    args.add_flag("--list", "List all kernel sets");
+
+    if(!args.parse(argc, argv))
+        return 0;
+
     std::cout << "======================================================================\n";
-    std::cout << "Example 04: Multi-Size Convolution with GPU Execution\n";
+    std::cout << "Example 03: Multi-Size Convolution with GPU Execution\n";
     std::cout << "======================================================================\n\n";
 
+    if(args.has("--list"))
+    {
+        std::cout << "Declared Kernel Sets:\n";
+        ConvKernelSetRegistry::instance().print();
+        return 0;
+    }
+
     // -------------------------------------------------------------------------
     // Step 1: Show declared kernels
     // -------------------------------------------------------------------------

dispatcher/examples/conv/cpp/05_heuristics.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
 
 /**
- * Example 06: Convolution Heuristics with GPU Execution
+ * Example 05: Convolution Heuristics with GPU Execution
  *
  * Demonstrates heuristic-based kernel selection with GPU execution.
  *
@@ -14,13 +14,15 @@
 #include <hip/hip_runtime.h>
 
 #include "ck_tile/dispatcher/conv_utils.hpp"
+#include "ck_tile/dispatcher/example_args.hpp"
 #include "ck_tile/core.hpp"
 #include "ck_tile/host.hpp"
 #include "ck_tile/host/convolution_parameter.hpp"
 #include "ck_tile/ops/grouped_convolution.hpp"
 
 using namespace ck_tile::dispatcher;
 using namespace ck_tile::dispatcher::conv_utils;
+using namespace ck_tile::dispatcher::utils;
 
 // =============================================================================
 // KERNEL DECLARATIONS
@@ -82,12 +84,25 @@ using OutDataType = ck_tile::half_t;
 // MAIN
 // =============================================================================
 
-int main()
+int main(int argc, char* argv[])
 {
+    ExampleArgs args("Example 05: Conv Heuristics", "Heuristic-based kernel selection");
+    args.add_flag("--list", "List all kernel sets");
+
+    if(!args.parse(argc, argv))
+        return 0;
+
     std::cout << "======================================================================\n";
-    std::cout << "Example 06: Convolution Heuristics with GPU Execution\n";
+    std::cout << "Example 05: Convolution Heuristics with GPU Execution\n";
     std::cout << "======================================================================\n\n";
 
+    if(args.has("--list"))
+    {
+        std::cout << "Declared Kernel Sets:\n";
+        ConvKernelSetRegistry::instance().print();
+        return 0;
+    }
+
     // -------------------------------------------------------------------------
     // Setup
     // -------------------------------------------------------------------------

dispatcher/examples/conv/cpp/06_json_export.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
 
 /**
- * Example 07: Convolution JSON Export with GPU Execution
+ * Example 06: Convolution JSON Export with GPU Execution
  *
  * Exports kernel configurations to JSON and runs on GPU.
  *
@@ -16,13 +16,15 @@
 #include <hip/hip_runtime.h>
 
 #include "ck_tile/dispatcher/conv_utils.hpp"
+#include "ck_tile/dispatcher/example_args.hpp"
 #include "ck_tile/core.hpp"
 #include "ck_tile/host.hpp"
 #include "ck_tile/host/convolution_parameter.hpp"
 #include "ck_tile/ops/grouped_convolution.hpp"
 
 using namespace ck_tile::dispatcher;
 using namespace ck_tile::dispatcher::conv_utils;
+using namespace ck_tile::dispatcher::utils;
 
 // =============================================================================
 // KERNEL DECLARATIONS
@@ -99,12 +101,26 @@ using OutDataType = ck_tile::half_t;
 // MAIN
 // =============================================================================
 
-int main()
+int main(int argc, char* argv[])
 {
+    ExampleArgs args("Example 06: Conv JSON Export", "Export kernel configurations to JSON");
+    args.add_option("--output", "conv_kernels.json", "Output JSON file path");
+    args.add_flag("--list", "List all kernel sets");
+
+    if(!args.parse(argc, argv))
+        return 0;
+
     std::cout << "======================================================================\n";
-    std::cout << "Example 07: Convolution JSON Export with GPU Execution\n";
+    std::cout << "Example 06: Convolution JSON Export with GPU Execution\n";
     std::cout << "======================================================================\n\n";
 
+    if(args.has("--list"))
+    {
+        std::cout << "Declared Kernel Sets:\n";
+        ConvKernelSetRegistry::instance().print();
+        return 0;
+    }
+
     // -------------------------------------------------------------------------
     // Export to JSON
     // -------------------------------------------------------------------------
@@ -117,12 +133,13 @@ int main()
     std::cout << json << "\n";
 
     // Write to file
-    std::ofstream file("conv_kernels.json");
+    std::string output_file = args.get("--output");
+    std::ofstream file(output_file);
     if(file)
     {
         file << json;
         file.close();
-        std::cout << "[Saved to conv_kernels.json]\n\n";
+        std::cout << "[Saved to " << output_file << "]\n\n";
     }
 
     // -------------------------------------------------------------------------

dispatcher/examples/conv/cpp/07_multi_registry.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
 
 /**
- * Example 08: Multiple Convolution Registries with GPU Execution
+ * Example 07: Multiple Convolution Registries with GPU Execution
  *
  * Demonstrates using separate registries for different use cases,
  * each running on GPU.
@@ -15,13 +15,15 @@
 #include <hip/hip_runtime.h>
 
 #include "ck_tile/dispatcher/conv_utils.hpp"
+#include "ck_tile/dispatcher/example_args.hpp"
 #include "ck_tile/core.hpp"
 #include "ck_tile/host.hpp"
 #include "ck_tile/host/convolution_parameter.hpp"
 #include "ck_tile/ops/grouped_convolution.hpp"
 
 using namespace ck_tile::dispatcher;
 using namespace ck_tile::dispatcher::conv_utils;
+using namespace ck_tile::dispatcher::utils;
 
 // =============================================================================
 // KERNEL DECLARATIONS - Different registries for different use cases
@@ -119,12 +121,26 @@ float run_conv(int N, int C, int K, int H, int W)
 // MAIN
 // =============================================================================
 
-int main()
+int main(int argc, char* argv[])
 {
+    ExampleArgs args("Example 07: Multi-Registry Conv",
+                     "Separate registries for different use cases");
+    args.add_flag("--list", "List all kernel sets");
+
+    if(!args.parse(argc, argv))
+        return 0;
+
     std::cout << "======================================================================\n";
-    std::cout << "Example 08: Multiple Convolution Registries with GPU Execution\n";
+    std::cout << "Example 07: Multiple Convolution Registries with GPU Execution\n";
     std::cout << "======================================================================\n\n";
 
+    if(args.has("--list"))
+    {
+        std::cout << "Declared Kernel Sets:\n";
+        ConvKernelSetRegistry::instance().print();
+        return 0;
+    }
+
     // -------------------------------------------------------------------------
     // Create separate registries
     // -------------------------------------------------------------------------

dispatcher/examples/conv/cpp/08_conv3d_forward.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
 
 /**
- * Example 09: 3D Convolution Forward with GPU Execution
+ * Example 08: 3D Convolution Forward with GPU Execution
  *
  * Demonstrates 3D convolution (e.g., for video or volumetric data).
  *
@@ -14,13 +14,15 @@
 #include <hip/hip_runtime.h>
 
 #include "ck_tile/dispatcher/conv_utils.hpp"
+#include "ck_tile/dispatcher/example_args.hpp"
 #include "ck_tile/core.hpp"
 #include "ck_tile/host.hpp"
 #include "ck_tile/host/convolution_parameter.hpp"
 #include "ck_tile/ops/grouped_convolution.hpp"
 
 using namespace ck_tile::dispatcher;
 using namespace ck_tile::dispatcher::conv_utils;
+using namespace ck_tile::dispatcher::utils;
 
 // =============================================================================
 // KERNEL DECLARATIONS - 3D Forward
@@ -56,12 +58,30 @@ using OutDataType = ck_tile::half_t;
 // MAIN
 // =============================================================================
 
-int main()
+int main(int argc, char* argv[])
 {
+    ExampleArgs args("Example 08: Conv3D Forward", "3D convolution for video/volumetric data");
+    args.add_option("-n", "1", "Batch size N");
+    args.add_option("-c", "32", "Input channels C");
+    args.add_option("-k", "64", "Output channels K");
+    args.add_option("--depth", "8", "Depth D");
+    args.add_option("--size", "16", "Spatial size (H=W)");
+    args.add_flag("--list", "List all kernel sets");
+
+    if(!args.parse(argc, argv))
+        return 0;
+
     std::cout << "======================================================================\n";
-    std::cout << "Example 09: 3D Convolution Forward with GPU Execution\n";
+    std::cout << "Example 08: 3D Convolution Forward with GPU Execution\n";
     std::cout << "======================================================================\n\n";
 
+    if(args.has("--list"))
+    {
+        std::cout << "Declared Kernel Sets:\n";
+        ConvKernelSetRegistry::instance().print();
+        return 0;
+    }
+
     // -------------------------------------------------------------------------
     // Step 1: Show declared kernels
     // -------------------------------------------------------------------------
@@ -78,9 +98,13 @@ int main()
     std::cout << "Step 2: Define 3D Problem\n";
     std::cout << "-------------------------\n";
 
-    // 3D problem: N=1, C=32, K=64, D=8, H=16, W=16, filter 3x3x3
-    int N = 1, C = 32, K = 64;
-    int Di = 8, Hi = 16, Wi = 16;
+    // 3D problem from args
+    int N  = args.get_int("-n", 1);
+    int C  = args.get_int("-c", 32);
+    int K  = args.get_int("-k", 64);
+    int Di = args.get_int("--depth", 8);
+    int Hi = args.get_int("--size", 16);
+    int Wi = Hi;
     int Z = 3, Y = 3, X = 3;
 
     auto problem = create_conv3d_problem(N, C, K, Di, Hi, Wi, Z, Y, X, 1, 1, ConvOp::Forward);