added wmma multiply_multiply instances

This commit is contained in:
Zoltan Lakatos
2025-06-17 15:03:18 +00:00
parent 881bc3fbb0
commit ac60286ed0
13 changed files with 360 additions and 60 deletions

View File

@@ -69,6 +69,8 @@ bool profile_gemm_multiply_multiply_impl(int do_verification,
}
};
std::cout << "cicc: " << StrideD0 << " " << StrideD1 << std::endl;
Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
Tensor<D0DataType> d0_m_n(f_host_tensor_descriptor(M, N, StrideD0, D0Layout{}));
@@ -97,8 +99,8 @@ bool profile_gemm_multiply_multiply_impl(int do_verification,
case 1:
a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-1, 2});
b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-1, 2});
d0_m_n.GenerateTensorValue(GeneratorTensor_2<D0DataType>{-5, 5});
d1_m_n.GenerateTensorValue(GeneratorTensor_2<D1DataType>{-1, 1});
d0_m_n.GenerateTensorValue(GeneratorTensor_2<D0DataType>{1, 3});
d1_m_n.GenerateTensorValue(GeneratorTensor_2<D1DataType>{1, 2});
break;
default:
a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});

View File

@@ -58,7 +58,6 @@ if(SUPPORTED_GPU_TARGETS MATCHES "gfx9")
endif()
list(APPEND PROFILER_OPS profile_gemm_multiply_add.cpp)
if(SUPPORTED_GPU_TARGETS MATCHES "gfx9[45]")
list(APPEND PROFILER_OPS profile_gemm_multiply_multiply.cpp)
list(APPEND PROFILER_OPS profile_gemm_multiply_multiply_wp.cpp)
list(APPEND PROFILER_OPS profile_gemm_ab_scale.cpp)
endif()
@@ -84,6 +83,9 @@ if((SUPPORTED_GPU_TARGETS MATCHES "gfx9" AND (DTYPES MATCHES "fp16" OR NOT DEFIN
(SUPPORTED_GPU_TARGETS MATCHES "gfx1[12]" AND (DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)))
list(APPEND PROFILER_OPS profile_gemm_bilinear.cpp)
endif()
#if((SUPPORTED_GPU_TARGETS MATCHES "gfx9[45]") OR (SUPPORTED_GPU_TARGETS MATCHES "gfx1[12]"))
list(APPEND PROFILER_OPS profile_gemm_multiply_multiply.cpp)
#endif()
if(SUPPORTED_GPU_TARGETS MATCHES "gfx9" OR SUPPORTED_GPU_TARGETS MATCHES "gfx1[12]")
list(APPEND PROFILER_OPS profile_gemm_universal.cpp)
@@ -149,7 +151,7 @@ if(SUPPORTED_GPU_TARGETS MATCHES "gfx9")
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
list(APPEND DEVICE_INSTANCES device_gemm_add_instance)
list(APPEND DEVICE_INSTANCES device_gemm_add_add_fastgelu_instance)
list(APPEND DEVICE_INSTANCES device_gemm_fastgelu_instance)
# list(APPEND DEVICE_INSTANCES device_gemm_fastgelu_instance)
list(APPEND DEVICE_INSTANCES device_batched_gemm_gemm_instance)
list(APPEND DEVICE_INSTANCES device_batched_gemm_add_relu_gemm_add_instance)
list(APPEND DEVICE_INSTANCES device_grouped_gemm_instance)
@@ -165,7 +167,6 @@ if(SUPPORTED_GPU_TARGETS MATCHES "gfx9")
list(APPEND DEVICE_INSTANCES device_batched_gemm_reduce_instance)
list(APPEND DEVICE_INSTANCES device_gemm_multiply_add_instance)
if(SUPPORTED_GPU_TARGETS MATCHES "gfx9[45]")
list(APPEND DEVICE_INSTANCES device_gemm_multiply_multiply_instance)
list(APPEND DEVICE_INSTANCES device_gemm_multiply_multiply_wp_instance)
list(APPEND DEVICE_INSTANCES device_gemm_ab_scale_instance)
endif()
@@ -195,6 +196,9 @@ if((SUPPORTED_GPU_TARGETS MATCHES "gfx9" AND (DTYPES MATCHES "fp16" OR NOT DEFIN
(SUPPORTED_GPU_TARGETS MATCHES "gfx1[12]" AND (DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)))
list(APPEND DEVICE_INSTANCES device_gemm_bilinear_instance)
endif()
#if((SUPPORTED_GPU_TARGETS MATCHES "gfx9[45]") OR (SUPPORTED_GPU_TARGETS MATCHES "gfx1[12]"))
list(APPEND DEVICE_INSTANCES device_gemm_multiply_multiply_instance)
#endif()
if(SUPPORTED_GPU_TARGETS MATCHES "gfx9" OR SUPPORTED_GPU_TARGETS MATCHES "gfx1[12]")
list(APPEND DEVICE_INSTANCES device_gemm_universal_instance)

View File

@@ -13,6 +13,8 @@ static void print_helper_message()
int main(int argc, char* argv[])
{
printf("cicc2\n");
if(argc == 1)
{
print_helper_message();