Merge branch 'develop' into codegen_hiprtc

Dockerfile

@@ -94,7 +94,7 @@ RUN pip install --upgrade cmake==3.27.5 && \
     dpkg -i dumb-init_*.deb && rm dumb-init_*.deb && \
 # Install packages for processing the performance results
     pip3 install --upgrade pip && \
-    pip3 install sqlalchemy==2.0.36 pymysql pandas==2.2.3 setuptools-rust sshtunnel==0.4.0 && \
+    pip3 install --upgrade pytest sqlalchemy==2.0.36 pymysql pandas==2.2.3 setuptools-rust setuptools>=75 sshtunnel==0.4.0 && \
 # Add render group
     groupadd -f render && \
 # Install the new rocm-cmake version

Jenkinsfile

@@ -512,6 +512,13 @@ def Build_CK(Map conf=[:]){
                         arch_type = 5
                     }
                     cmake_build(conf)
+                    if ( arch_type == 1 ){
+                            echo "Run inductor codegen tests"
+                            sh """
+                                  pip install --verbose .
+                                  pytest python/test/test_gen_instances.py
+                            """
+                    }
                     dir("build"){
                         if (params.RUN_FULL_QA && arch_type == 1 ){
                             // build deb packages for all gfx9 targets on gfx90a system and prepare to export

library/include/ck/library/tensor_operation_instance/gpu/gemm_multiply_multiply.hpp

@@ -16,7 +16,8 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_FP8))
+#ifdef CK_ENABLE_FP8
+#ifdef CK_ENABLE_BF16
 void add_device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_comp_default_instances(
     std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
                                                           Col,
@@ -95,6 +96,86 @@ void add_device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_mem_v2_kpadding_i
                                                           PassThrough,
                                                           MultiplyMultiply>>>& instances);
 #endif
+#ifdef CK_ENABLE_FP16
+void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
+                                                          Col,
+                                                          Tuple<Row, Col>,
+                                                          Row,
+                                                          F8,
+                                                          F8,
+                                                          Tuple<F32, F32>,
+                                                          F16,
+                                                          PassThrough,
+                                                          PassThrough,
+                                                          MultiplyMultiply>>>& instances);
+
+void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
+                                                          Col,
+                                                          Tuple<Row, Col>,
+                                                          Row,
+                                                          F8,
+                                                          F8,
+                                                          Tuple<F32, F32>,
+                                                          F16,
+                                                          PassThrough,
+                                                          PassThrough,
+                                                          MultiplyMultiply>>>& instances);
+
+void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v1_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
+                                                          Col,
+                                                          Tuple<Row, Col>,
+                                                          Row,
+                                                          F8,
+                                                          F8,
+                                                          Tuple<F32, F32>,
+                                                          F16,
+                                                          PassThrough,
+                                                          PassThrough,
+                                                          MultiplyMultiply>>>& instances);
+
+void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v1_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
+                                                          Col,
+                                                          Tuple<Row, Col>,
+                                                          Row,
+                                                          F8,
+                                                          F8,
+                                                          Tuple<F32, F32>,
+                                                          F16,
+                                                          PassThrough,
+                                                          PassThrough,
+                                                          MultiplyMultiply>>>& instances);
+
+void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v2_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
+                                                          Col,
+                                                          Tuple<Row, Col>,
+                                                          Row,
+                                                          F8,
+                                                          F8,
+                                                          Tuple<F32, F32>,
+                                                          F16,
+                                                          PassThrough,
+                                                          PassThrough,
+                                                          MultiplyMultiply>>>& instances);
+
+void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v2_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
+                                                          Col,
+                                                          Tuple<Row, Col>,
+                                                          Row,
+                                                          F8,
+                                                          F8,
+                                                          Tuple<F32, F32>,
+                                                          F16,
+                                                          PassThrough,
+                                                          PassThrough,
+                                                          MultiplyMultiply>>>& instances);
+#endif
+#endif
 
 #if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_INT8))
 void add_device_gemm_multiply_multiply_xdl_i8_i8_bf16_mk_nk_mn_comp_default_instances(
@@ -213,7 +294,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
-#if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_FP8))
+#ifdef CK_ENABLE_FP8
+#ifdef CK_ENABLE_BF16
         if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, f8_t> &&
                      is_same_v<CDataType, bhalf_t>)
         {
@@ -237,6 +319,31 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
             }
         }
 #endif
+#ifdef CK_ENABLE_FP16
+        if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, f8_t> &&
+                     is_same_v<CDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                         is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_default_instances(
+                    op_ptrs);
+                add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_kpadding_instances(
+                    op_ptrs);
+
+                add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v1_default_instances(
+                    op_ptrs);
+                add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v1_kpadding_instances(
+                    op_ptrs);
+
+                add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v2_default_instances(
+                    op_ptrs);
+                add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v2_kpadding_instances(
+                    op_ptrs);
+            }
+        }
+#endif
+#endif
 #if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_INT8))
         if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
                      is_same_v<CDataType, bhalf_t>)

library/src/tensor_operation_instance/gpu/gemm_multiply_multiply/CMakeLists.txt

@@ -9,6 +9,13 @@ list(APPEND GEMM_MULTIPLY_MULTIPLY_INSTANCES
         device_gemm_multiply_multiply_xdl_f8_f8_bf16/device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_mem_v2_default_instance.cpp
         device_gemm_multiply_multiply_xdl_f8_f8_bf16/device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_mem_v2_kpadding_instance.cpp
 
+        device_gemm_multiply_multiply_xdl_f8_f8_f16/device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_default_instance.cpp
+        device_gemm_multiply_multiply_xdl_f8_f8_f16/device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_kpadding_instance.cpp
+        device_gemm_multiply_multiply_xdl_f8_f8_f16/device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v1_default_instance.cpp
+        device_gemm_multiply_multiply_xdl_f8_f8_f16/device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v1_kpadding_instance.cpp
+        device_gemm_multiply_multiply_xdl_f8_f8_f16/device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v2_default_instance.cpp
+        device_gemm_multiply_multiply_xdl_f8_f8_f16/device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v2_kpadding_instance.cpp
+
         device_gemm_multiply_multiply_xdl_i8_i8_bf16/device_gemm_multiply_multiply_xdl_i8_i8_bf16_mk_nk_mn_comp_default_instance.cpp
         device_gemm_multiply_multiply_xdl_i8_i8_bf16/device_gemm_multiply_multiply_xdl_i8_i8_bf16_mk_nk_mn_comp_kpadding_instance.cpp
         device_gemm_multiply_multiply_xdl_i8_i8_bf16/device_gemm_multiply_multiply_xdl_i8_i8_bf16_mk_nk_mn_mem_v1_default_instance.cpp
@@ -20,6 +27,9 @@ list(APPEND GEMM_MULTIPLY_MULTIPLY_INSTANCES
 set_source_files_properties(device_gemm_multiply_multiply_xdl_f8_f8_bf16/device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_comp_default_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")
 set_source_files_properties(device_gemm_multiply_multiply_xdl_f8_f8_bf16/device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_comp_kpadding_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")
 
+set_source_files_properties(device_gemm_multiply_multiply_xdl_f8_f8_f16/device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_default_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")
+set_source_files_properties(device_gemm_multiply_multiply_xdl_f8_f8_f16/device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_kpadding_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")
+
 set_source_files_properties(device_gemm_multiply_multiply_xdl_i8_i8_bf16/device_gemm_multiply_multiply_xdl_i8_i8_bf16_mk_nk_mn_comp_default_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")
 set_source_files_properties(device_gemm_multiply_multiply_xdl_i8_i8_bf16/device_gemm_multiply_multiply_xdl_i8_i8_bf16_mk_nk_mn_comp_kpadding_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")
 

library/src/tensor_operation_instance/gpu/gemm_multiply_multiply/device_gemm_multiply_multiply_xdl_f8_f8_f16/device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_default_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
+                                                          Col,
+                                                          Tuple<Row, Col>,
+                                                          Row,
+                                                          F8,
+                                                          F8,
+                                                          Tuple<F32, F32>,
+                                                          F16,
+                                                          PassThrough,
+                                                          PassThrough,
+                                                          MultiplyMultiply>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_instances<GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_multiply_multiply/device_gemm_multiply_multiply_xdl_f8_f8_f16/device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_kpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
+                                                          Col,
+                                                          Tuple<Row, Col>,
+                                                          Row,
+                                                          F8,
+                                                          F8,
+                                                          Tuple<F32, F32>,
+                                                          F16,
+                                                          PassThrough,
+                                                          PassThrough,
+                                                          MultiplyMultiply>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_comp_instances<GemmKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_multiply_multiply/device_gemm_multiply_multiply_xdl_f8_f8_f16/device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v1_default_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v1_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
+                                                          Col,
+                                                          Tuple<Row, Col>,
+                                                          Row,
+                                                          F8,
+                                                          F8,
+                                                          Tuple<F32, F32>,
+                                                          F16,
+                                                          PassThrough,
+                                                          PassThrough,
+                                                          MultiplyMultiply>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_instances<Intrawave,
+                                                                           GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_multiply_multiply/device_gemm_multiply_multiply_xdl_f8_f8_f16/device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v1_kpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v1_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
+                                                          Col,
+                                                          Tuple<Row, Col>,
+                                                          Row,
+                                                          F8,
+                                                          F8,
+                                                          Tuple<F32, F32>,
+                                                          F16,
+                                                          PassThrough,
+                                                          PassThrough,
+                                                          MultiplyMultiply>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_instances<Intrawave,
+                                                                           GemmKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_multiply_multiply/device_gemm_multiply_multiply_xdl_f8_f8_f16/device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v2_default_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v2_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
+                                                          Col,
+                                                          Tuple<Row, Col>,
+                                                          Row,
+                                                          F8,
+                                                          F8,
+                                                          Tuple<F32, F32>,
+                                                          F16,
+                                                          PassThrough,
+                                                          PassThrough,
+                                                          MultiplyMultiply>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_instances<Interwave,
+                                                                           GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_multiply_multiply/device_gemm_multiply_multiply_xdl_f8_f8_f16/device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v2_kpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_v2_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDSplitK<Row,
+                                                          Col,
+                                                          Tuple<Row, Col>,
+                                                          Row,
+                                                          F8,
+                                                          F8,
+                                                          Tuple<F32, F32>,
+                                                          F16,
+                                                          PassThrough,
+                                                          PassThrough,
+                                                          MultiplyMultiply>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_multiply_multiply_xdl_f8_f8_f16_mk_nk_mn_mem_instances<Interwave,
+                                                                           GemmKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

profiler/src/profile_gemm_multiply_multiply.cpp

@@ -28,6 +28,7 @@ enum struct GemmDataType
     F16_F16_F16_F8, // 6
     F8_F8_BF16,     // 7
     INT8_INT8_BF16, // 8
+    F8_F8_F16,      // 9
 };
 
 #define OP_NAME "gemm_multiply_multiply"
@@ -40,7 +41,7 @@ int profile_gemm_multiply_multiply(int argc, char* argv[])
         printf("arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n");
         printf("arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8; 4: f8@f16; 5: f16@f8; 6: "
                "f16->f8; 7: f8->bf16, "
-               "comp f8; 8: int8->bf16)\n");
+               "comp f8; 8: int8->bf16; 9: f8->f16, comp f8;)\n");
         printf("arg3: matrix layout (0: A[m, k] * B[k, n] = C[m, n];\n");
         printf("                     1: A[m, k] * B[n, k] = C[m, n];\n");
         printf("                     2: A[k, m] * B[k, n] = C[m, n];\n");
@@ -89,6 +90,7 @@ int profile_gemm_multiply_multiply(int argc, char* argv[])
 
     using F32  = float;
     using BF16 = ck::bhalf_t;
+    using F16  = ck::half_t;
     using F8   = ck::f8_t;
     using I8   = int8_t;
     using I32  = int;
@@ -165,6 +167,11 @@ int profile_gemm_multiply_multiply(int argc, char* argv[])
         return profile(
             F8{}, F8{}, F8{}, F32{}, F32{}, F32{}, BF16{}, Row{}, Col{}, Row{}, Col{}, Row{});
     }
+    else if(data_type == GemmDataType::F8_F8_F16 && layout == GemmMatrixLayout::MK_NK_MN)
+    {
+        return profile(
+            F8{}, F8{}, F8{}, F32{}, F32{}, F32{}, F16{}, Row{}, Col{}, Row{}, Col{}, Row{});
+    }
     else if(data_type == GemmDataType::INT8_INT8_BF16 && layout == GemmMatrixLayout::MK_NK_MN)
     {
         return profile(

pyproject.toml

@@ -21,16 +21,19 @@ dependencies = []
 "Bug Tracker" = "https://github.com/rocm/composable_kernel/issues"
 
 [tool.setuptools]
-packages = ["ck4inductor", "ck4inductor.include", "ck4inductor.library"]
+packages = ["ck4inductor", "ck4inductor.include", "ck4inductor.library", "ck4inductor.universal_gemm", "ck4inductor.batched_universal_gemm", "ck4inductor.grouped_conv_fwd"]
 
 [tool.setuptools.package-dir]
 ck4inductor = "python/ck4inductor"
+"ck4inductor.universal_gemm" = "python/ck4inductor/universal_gemm"
+"ck4inductor.batched_universal_gemm" = "python/ck4inductor/batched_universal_gemm"
+"ck4inductor.grouped_conv_fwd" = "python/ck4inductor/grouped_conv_fwd"
 "ck4inductor.include" = "include"
 "ck4inductor.library" = "library"
 
 [tool.setuptools.package-data]
 "ck4inductor.include" = ["ck/**/*.hpp"]
-"ck4inductor.library" = ["src/tensor_operation_instance/gpu/gemm_universal/**/*.hpp"]
+"ck4inductor.library" = ["src/tensor_operation_instance/gpu/gemm_universal/**/*.hpp", "src/tensor_operation_instance/gpu/gemm_universal_batched/**/*.hpp", "include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/**/*.hpp"]
 
 [tool.setuptools.dynamic]
 version = { attr = "setuptools_scm.get_version" }

python/ck4inductor/universal_gemm/gen_instances.py

@@ -68,12 +68,13 @@ def parse_instances(str_instances: List[str]) -> List[CKGemmOperation]:
 
         template_args.insert(2, tuple())  # ds layout
         template_args.insert(6, tuple())  # ds dtype
-
-        new_instance = CKGemmOperation(
-            *template_args,  # type: ignore[arg-type]
-        )
-
-        op_instances.append(new_instance)
+        try:
+            new_instance = CKGemmOperation(
+                *template_args,  # type: ignore[arg-type]
+            )
+            op_instances.append(new_instance)
+        except TypeError as e:
+            log.debug(f"{e} when parsing {line}")
     return op_instances
 
 

python/test/test_gen_instances.py

+# SPDX-License-Identifier: MIT
+# Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+import logging
+
+import unittest
+
+from ck4inductor.universal_gemm.gen_instances import (
+    gen_ops_library as gen_gemm_ops_library,
+)
+from ck4inductor.universal_gemm.gen_instances import (
+    gen_ops_preselected as gen_gemm_ops_preselected,
+)
+from ck4inductor.grouped_conv_fwd.gen_instances import (
+    gen_conv_ops_library as gen_conv_ops_library,
+)
+from ck4inductor.batched_universal_gemm.gen_instances import (
+    gen_ops_library as gen_batched_gemm_ops_library,
+)
+
+log = logging.getLogger(__name__)
+
+
+class TestGenInstances(unittest.TestCase):
+    def test_gen_gemm_instances(self):
+        instances = gen_gemm_ops_library()
+
+        log.debug("%d gemm instances from library" % len(instances))
+        self.assertTrue(instances)
+
+    def test_preselected_gemm_instances(self):
+        instances = gen_gemm_ops_preselected()
+
+        log.debug("%d preselected gemm instances" % len(instances))
+        self.assertTrue(instances)
+
+    def test_gen_conv_instances(self):
+        instances = gen_conv_ops_library()
+
+        log.debug("%d gemm instances from library" % len(instances))
+        self.assertTrue(instances)
+
+    def test_gen_batched_gemm_instances(self):
+        instances = gen_batched_gemm_ops_library()
+
+        log.debug("%d gemm instances from library" % len(instances))
+        self.assertTrue(instances)