From f3e08fa9d68d1a434678897c63cfd5a84a59f9b8 Mon Sep 17 00:00:00 2001 From: Matthias Gehre Date: Thu, 27 Nov 2025 08:48:00 +0100 Subject: [PATCH] Add support for gfx1153 (#3306) [ROCm/composable_kernel commit: 678298d4c7141d41a552e7d8fea396ee88a4652f] --- .../65_gemm_multiply_multiply/CMakeLists.txt | 2 +- example/CMakeLists.txt | 2 +- include/ck/ck.hpp | 2 +- include/ck/host_utility/device_prop.hpp | 2 +- include/ck_tile/core/config.hpp | 2 +- include/ck_tile/host/device_prop.hpp | 2 +- .../gpu/CMakeLists.txt | 20 +++++++++---------- test/CMakeLists.txt | 6 +++--- 8 files changed, 19 insertions(+), 19 deletions(-) diff --git a/example/65_gemm_multiply_multiply/CMakeLists.txt b/example/65_gemm_multiply_multiply/CMakeLists.txt index 74930d2b21..b46b606ed4 100644 --- a/example/65_gemm_multiply_multiply/CMakeLists.txt +++ b/example/65_gemm_multiply_multiply/CMakeLists.txt @@ -16,7 +16,7 @@ add_example_executable(example_moe_gemm2_xdl_fp8 moe_gemm2_xdl_fp8.cpp) add_example_executable(example_moe_gemm2_xdl_fp8_blockscale moe_gemm2_xdl_fp8_blockscale.cpp) add_example_executable(example_moe_gemm1_xdl_fp8_blockscale moe_gemm1_xdl_fp8_blockscale.cpp) -list(APPEND gpu_list gfx942 gfx950 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1200 gfx1201 gfx11-generic gfx12-generic) +list(APPEND gpu_list gfx942 gfx950 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx1200 gfx1201 gfx11-generic gfx12-generic) set(target 0) foreach(gpu IN LISTS GPU_TARGETS) if(gpu IN_LIST gpu_list AND target EQUAL 0) diff --git a/example/CMakeLists.txt b/example/CMakeLists.txt index 940e7bc5e6..579c5a9759 100644 --- a/example/CMakeLists.txt +++ b/example/CMakeLists.txt @@ -113,7 +113,7 @@ function(add_example_executable EXAMPLE_NAME FILE_NAME) elseif(source_name_list MATCHES "_wmma") list(REMOVE_ITEM EX_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack+ gfx908:xnack- gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx942 gfx1030 gfx950) elseif(source_name_list MATCHES "_mx") #only build mx example for gfx950 - list(REMOVE_ITEM EX_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack+ gfx908:xnack- gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx942 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1200 gfx1201 gfx10-3-generic gfx11-generic gfx12-generic) + list(REMOVE_ITEM EX_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack+ gfx908:xnack- gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx942 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx1200 gfx1201 gfx10-3-generic gfx11-generic gfx12-generic) elseif(source_name_list MATCHES "_pk_i4") #only build these examples for gfx942 gfx950 and rdna3/4 message(DEBUG "trimming targets for ${FILE_NAME}") list(REMOVE_ITEM EX_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack+ gfx908:xnack- gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx1030 gfx10-3-generic) diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp index 9debdc12b2..879fb31ca5 100644 --- a/include/ck/ck.hpp +++ b/include/ck/ck.hpp @@ -69,7 +69,7 @@ #endif #if defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__) || \ defined(__gfx1103__) || defined(__gfx1150__) || defined(__gfx1151__) || \ - defined(__gfx1152__) || defined(__gfx11_generic__) + defined(__gfx1152__) || defined(__gfx1153__) || defined(__gfx11_generic__) #define __gfx11__ #endif #if defined(__gfx1200__) || defined(__gfx1201__) || defined(__gfx12_generic__) diff --git a/include/ck/host_utility/device_prop.hpp b/include/ck/host_utility/device_prop.hpp index 6db5c0a071..8739f65740 100644 --- a/include/ck/host_utility/device_prop.hpp +++ b/include/ck/host_utility/device_prop.hpp @@ -62,7 +62,7 @@ inline bool is_gfx11_supported() return ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" || ck::get_device_name() == "gfx1102" || ck::get_device_name() == "gfx1103" || ck::get_device_name() == "gfx1150" || ck::get_device_name() == "gfx1151" || - ck::get_device_name() == "gfx1152"; + ck::get_device_name() == "gfx1152" || ck::get_device_name() == "gfx1153"; } inline bool is_xdl_supported() diff --git a/include/ck_tile/core/config.hpp b/include/ck_tile/core/config.hpp index 776a9fa01c..de97b46336 100644 --- a/include/ck_tile/core/config.hpp +++ b/include/ck_tile/core/config.hpp @@ -21,7 +21,7 @@ #endif #if defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__) || \ defined(__gfx1103__) || defined(__gfx1150__) || defined(__gfx1151__) || \ - defined(__gfx1152__) || defined(__gfx11_generic__) + defined(__gfx1152__) || defined(__gfx1153__) || defined(__gfx11_generic__) #define __gfx11__ #endif #if defined(__gfx1200__) || defined(__gfx1201__) || defined(__gfx12_generic__) diff --git a/include/ck_tile/host/device_prop.hpp b/include/ck_tile/host/device_prop.hpp index bd01fa8d88..2d7dc7dd18 100644 --- a/include/ck_tile/host/device_prop.hpp +++ b/include/ck_tile/host/device_prop.hpp @@ -57,7 +57,7 @@ inline bool is_gfx11_supported() return get_device_name() == "gfx1100" || get_device_name() == "gfx1101" || get_device_name() == "gfx1102" || get_device_name() == "gfx1103" || get_device_name() == "gfx1150" || get_device_name() == "gfx1151" || - get_device_name() == "gfx1152"; + get_device_name() == "gfx1152" || get_device_name() == "gfx1153"; } inline bool is_gfx12_supported() diff --git a/library/src/tensor_operation_instance/gpu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/CMakeLists.txt index 6f171191ca..98db603313 100644 --- a/library/src/tensor_operation_instance/gpu/CMakeLists.txt +++ b/library/src/tensor_operation_instance/gpu/CMakeLists.txt @@ -115,39 +115,39 @@ function(add_instance_library INSTANCE_NAME) elseif(source_name MATCHES "_wmma") list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack+ gfx908:xnack- gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx942 gfx1030 gfx950) elseif(source_name MATCHES "mha") - list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx908 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1200 gfx1201 gfx10-3-generic gfx11-generic gfx12-generic) + list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx908 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx1200 gfx1201 gfx10-3-generic gfx11-generic gfx12-generic) endif() if(source_name MATCHES "_mx") - list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx942 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1200 gfx1201 gfx10-3-generic gfx11-generic gfx12-generic) + list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx942 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx1200 gfx1201 gfx10-3-generic gfx11-generic gfx12-generic) endif() #only build the fp8 gemm instances for gfx90a if the build argument is set, otherwise only build for gfx942/gfx950 and gfx1200/gfx1201 if(NOT CK_USE_FP8_ON_UNSUPPORTED_ARCH) if(source_name MATCHES "gemm_xdl_universal" AND source_name MATCHES "f8") - list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx10-3-generic gfx11-generic) + list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx10-3-generic gfx11-generic) endif() if(source_name MATCHES "gemm_multiply_multiply" AND source_name MATCHES "f8") - list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx10-3-generic gfx11-generic) + list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx10-3-generic gfx11-generic) endif() if(source_name MATCHES "gemm_universal_preshuffle" AND source_name MATCHES "f8") - list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx10-3-generic gfx11-generic) + list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx10-3-generic gfx11-generic) endif() if(source_name MATCHES "gemm_xdl_universal_preshuffle" AND source_name MATCHES "f8") - list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx10-3-generic gfx11-generic) + list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx10-3-generic gfx11-generic) endif() else() if(source_name MATCHES "gemm_xdl_universal" AND source_name MATCHES "f8") - list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx908 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx10-3-generic gfx11-generic) + list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx908 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx10-3-generic gfx11-generic) endif() if(source_name MATCHES "gemm_multiply_multiply" AND source_name MATCHES "f8") - list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx908 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx10-3-generic gfx11-generic) + list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx908 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx10-3-generic gfx11-generic) endif() if(source_name MATCHES "gemm_universal_preshuffle" AND source_name MATCHES "f8") - list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx908 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx10-3-generic gfx11-generic) + list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx908 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx10-3-generic gfx11-generic) endif() if(source_name MATCHES "gemm_xdl_universal_preshuffle" AND source_name MATCHES "f8") - list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx908 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx10-3-generic gfx11-generic) + list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx908 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx10-3-generic gfx11-generic) endif() endif() if(source_name MATCHES "gemm_wmma_universal" AND source_name MATCHES "f8") diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt index 84c2ea090b..b00cd7789e 100644 --- a/test/CMakeLists.txt +++ b/test/CMakeLists.txt @@ -117,7 +117,7 @@ function(add_test_executable TEST_NAME) elseif(source_name_list MATCHES "_wmma") list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack+ gfx908:xnack- gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx942 gfx1030 gfx950) elseif(source_name_list MATCHES "_smfmac") - list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx906:xnack- gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx908 gfx90a gfx1200 gfx1201 gfx10-3-generic gfx11-generic gfx12-generic) + list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx906:xnack- gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx908 gfx90a gfx1200 gfx1201 gfx10-3-generic gfx11-generic gfx12-generic) endif() set_source_files_properties(${ARGN} PROPERTIES LANGUAGE HIP) add_executable(${TEST_NAME} ${ARGN}) @@ -209,9 +209,9 @@ function(add_gtest_executable TEST_NAME) elseif(source_name_list MATCHES "_wmma") list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack+ gfx908:xnack- gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx942 gfx1030 gfx950) elseif(source_name_list MATCHES "_smfmac") - list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx906:xnack- gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx908 gfx90a gfx1200 gfx1201 gfx10-3-generic gfx11-generic gfx12-generic) + list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx906:xnack- gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx908 gfx90a gfx1200 gfx1201 gfx10-3-generic gfx11-generic gfx12-generic) elseif(source_name_list MATCHES "_mx") #only build mx example for gfx950 - list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack+ gfx908:xnack- gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx942 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1200 gfx1201 gfx10-3-generic gfx11-generic gfx12-generic) + list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack+ gfx908:xnack- gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx942 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx1200 gfx1201 gfx10-3-generic gfx11-generic gfx12-generic) endif() set_source_files_properties(${ARGN} PROPERTIES LANGUAGE HIP) add_executable(${TEST_NAME} ${ARGN})