Code updated according to the PR comments.

Signed-off-by: Yasushi Negishi <[email protected]>

Author: negiyas

src/Builder/FrontendDialectTransformer.cpp

@@ -453,10 +453,9 @@ class FrontendGenImpl {
   }
 
   // Generate a string vector from the dimParams option string
-  void getInputDimParamsVecFromOption(std::string optionStr, size_t numOfArgs,
-      SmallVector<std::string> &paramStrVec) {
-    std::stringstream paramStrStream;
-    paramStrStream << optionStr;
+  void getInputDimParamsVecFromOption(std::string optionStr,
+      SmallVector<std::string> &paramStrVec, std::string &paramStrForAllArgs) {
+    std::stringstream paramStrStream(optionStr);
     std::string dimParamStr;
     while (std::getline(paramStrStream, dimParamStr, '|')) {
       size_t pos = dimParamStr.find(':');
@@ -465,10 +464,12 @@ class FrontendGenImpl {
       dimParamStr = dimParamStr.substr(pos + 1);
       std::replace(dimParamStr.begin(), dimParamStr.end(), '=', ':');
       if (idx < 0) // set all arguments
-        for (size_t i = 0; i < numOfArgs; i++)
-          paramStrVec[i] = dimParamStr;
-      else
+        paramStrForAllArgs = dimParamStr;
+      else {
+        while ((int)paramStrVec.size() <= idx) // Expand paramStrVec
+          paramStrVec.emplace_back("");
         paramStrVec[idx] = dimParamStr;
+      }
     }
     return;
   }
@@ -513,10 +514,10 @@ class FrontendGenImpl {
     // See https://github.com/onnx/onnx/blob/main/docs/IR.md for more
     // information about dim_param.
     llvm::SmallVector<std::string, 4> inputDimParams, outputDimParams;
-    size_t numOfArgs = graph.input().size();
-    llvm::SmallVector<std::string> inputDimParamsFromOption(numOfArgs);
-    getInputDimParamsVecFromOption(
-        options_.dimParams, numOfArgs, inputDimParamsFromOption);
+    llvm::SmallVector<std::string> inputDimParamsFromOption;
+    std::string inputDimParamsFromOptionForAllArgs;
+    getInputDimParamsVecFromOption(options_.dimParams, inputDimParamsFromOption,
+        inputDimParamsFromOptionForAllArgs);
 
     // Import the input tensor types that are not constant and not initialized.
     int inputIndex = 0;
@@ -527,8 +528,11 @@ class FrontendGenImpl {
         std::string dimParams = "";
         Type argTy = ImportType(input.type(), &dimParams);
         argTy = modelInputShaper_.reshape(inputIndex, argTy);
-        if (!inputDimParamsFromOption[inputIndex].empty())
+        if (inputIndex < (int)inputDimParamsFromOption.size() &&
+            !inputDimParamsFromOption[inputIndex].empty())
           inputDimParams.emplace_back(inputDimParamsFromOption[inputIndex]);
+        else if (!inputDimParamsFromOptionForAllArgs.empty())
+          inputDimParams.emplace_back(inputDimParamsFromOptionForAllArgs);
         else if (!dimParams.empty())
           inputDimParams.emplace_back(dimParams);
 

src/Builder/FrontendDialectTransformer.hpp

@@ -56,12 +56,14 @@ struct ImportOptions {
   //
   std::string shapeInformation = "";
   // Custom onnx.dim_params attributes for the graph inputs for specifying
-  // relationship among their ranks.
-  // Its format is 'input_id:dim=sym,dim=sym,dim=sym|input_id:dim=sym,dim=sym..'
+  // relationship among their dynamic dimensions.
+  // Its format is 'input_id:dim_id=sym,dim_id=sym,...|input_id:
+  // dim_id=sym,dim_id=sym,...|input_id...'
   // E.g. An ONNX model has two dynamic inputs
   //   - (arg0: tensor<?x5xf32>, arg1: tensor<?x5xf32>)
-  // If we want to specify that the first ranks of the first and second
-  // arguments are the same, we can use:
+  // If we want to specify that the first unknown dimension of arg0 and the
+  // first unknown dimension of arg1 are the same, we can assign the two
+  // dimensions to the same symbol "batch" as follows.
   //   - dimParams = '0:0=batch|1:0=batch'
   //
   std::string dimParams = "";

src/Compiler/CompilerOptions.cpp

@@ -285,11 +285,11 @@ static llvm::cl::opt<std::string, true> shapeInformationOpt("shapeInformation",
 static llvm::cl::opt<std::string, true> dimParamsOpt("dimParams",
     llvm::cl::desc(
         "Custom onnx.dim_params attributes for the inputs of the ONNX model for"
-        "specifying relationship among ranks of the inputs.\n"
+        "specifying relationship among dynamic dimensions of the inputs.\n"
         "\"value\" is in the format of "
-        "\"INPUT_ID1:D1=S1,D2=S2,...,Dn=Sn|INPUT_ID2:D1=T1,D2=T2,...Dn=Tn\""
-        "where \"INPUT_ID1, INPUT_ID2, ...\" are input indices (starting from "
-        "0 or being -1 for all input indices), and\n"
+        "\"INPUT_ID1:D1=S1,D2=S2,...,Dn=Sn|INPUT_ID2:D1=T1,D2=T2,...Dn=Tn|"
+        "...\" where \"INPUT_ID1, INPUT_ID2, ...\" are input indices "
+        "(starting from 0 or being -1 for all input indices), and\n"
         "\"S1, S2, ...\" and \"T2, T2, ...\" are symbols to specify that same "
         "symbols have the same value"),
     llvm::cl::value_desc("value"), llvm::cl::location(dimParams),

test/accelerators/NNPA/backend/CMakeLists.txt

@@ -100,8 +100,7 @@ endif()
 # "zdnn_add" to check the function is called, (3) when dimParams is
 # "NO_DYNAMIC_SHAPE_TEST", backend test is skipped, otherwise the string is
 # passed as --dimParams option. "0:0=a,1=b,2=c|1:0=a,1=b,2=c" means that the
-# first ranks of the first, second and third ranks of the first and second
-# input arguments are the same respectively.
+# first, second and third dimensions of the first and second input arguments # are the same respectively.
 set(NNPA_TEST_LIST
 
     # ==ARCH== NNPA

test/backend/common.py

@@ -67,13 +67,6 @@ def execute_commands(cmds, dynamic_inputs_dims):
     subprocess.run(cmds, env=my_env, check=True)
 
 
-def get_compile_option(test_name):
-    if args.dynamic and test_name in variables.test_to_enable_dimparams_dict:
-        return ["--dimParams=" + variables.test_to_enable_dimparams_dict[test_name]]
-    else:
-        return []
-
-
 def check_instruction(test_name, exec_name):
     if args.instruction_check and test_name in variables.test_to_enable_symbol_dict:
         symbol_name = variables.test_to_enable_symbol_dict[test_name]
@@ -115,6 +108,7 @@ def compile_model(model, emit):
 
     exec_base = os.path.join(model_dir, name)
     exec_name = exec_base + suffix[emit]
+    test_name_cpu = name + "_cpu"
 
     # Command
     command_list = [TEST_DRIVER]
@@ -139,7 +133,10 @@ def compile_model(model, emit):
             "--constants-to-file-total-threshold="
             + str(args.constants_to_file_total_threshold)
         )
-    command_list += get_compile_option(name + "_cpu")
+    if args.dynamic and test_name_cpu in variables.test_to_enable_dimparams_dict:
+        command_list.append(
+            "--dimParams=" + variables.test_to_enable_dimparams_dict[test_name_cpu]
+        )
 
     command_list.append(target[emit])
     command_list.append(model_name)
@@ -161,6 +158,6 @@ def compile_model(model, emit):
     # in execute_commands when calling subprocess.run.
 
     # Check if specific instruction are included in the compiled model.
-    check_instruction(name + "_cpu", exec_name)
+    check_instruction(test_name_cpu, exec_name)
 
     return exec_name