已合并
A5A6AIV兼容性适配 #2085
shirokawa_创建于 7月21日
A5A6AIV兼容性适配 #2085
已合并
shirokawa_创建于 7月21日
17 个文件变更+1569-3
MCMakeLists.txt+2-0文件内容审核中,请稍后刷新重试
Mbuild.sh+9-2
@@ -26,6 +26,12 @@ CUSTOM_OPTION="-DCMAKE_INSTALL_PREFIX=${OUTPUT_DIR}"
26STATIC_MODE="false" # 新增变量,用于控制是否静态编译26STATIC_MODE="false" # 新增变量,用于控制是否静态编译
27ENABLE_BUILD_DEVICE="OFF"27ENABLE_BUILD_DEVICE="OFF"
28ENABLE_BUILD_AARCH="OFF" 28ENABLE_BUILD_AARCH="OFF"
29+# 910_96(960) AIV kernel 编译开关:ENABLE_910_96="ON"即编, ENABLE_910_96="OFF"即不编
30+# 如需覆盖,请编辑同目录下的 build_config.sh,无需改动本文件
31+ENABLE_910_96="OFF"
32+if [ -f "${CURRENT_DIR}/build_config.sh" ]; then
33+ . "${CURRENT_DIR}/build_config.sh"
34+fi
29CANN_3RD_LIB_PATH="${CURRENT_DIR}/third_party"35CANN_3RD_LIB_PATH="${CURRENT_DIR}/third_party"
30CUSTOM_SIGN_SCRIPT=""36CUSTOM_SIGN_SCRIPT=""
31ENABLE_SIGN="false"37ENABLE_SIGN="false"
@@ -303,8 +309,8 @@ function build_static() {
303 rm -f "${EXTRACT_DIR}/${AIV_STEM}.bin" 309 rm -f "${EXTRACT_DIR}/${AIV_STEM}.bin"
304 AIV_EMBED_COUNT=$((AIV_EMBED_COUNT + 1)) 310 AIV_EMBED_COUNT=$((AIV_EMBED_COUNT + 1))
305 log "Info: Embedded AIV kernel: ${AIV_BASENAME}" 311 log "Info: Embedded AIV kernel: ${AIV_BASENAME}"
306- done < <(find "${BUILD_DIR}/src/ops" -type f \ 312+ done < <(find "${BUILD_DIR}/src/ops" -type f \
307- -name 'hccl_aiv_*_op_910_95.o' -print0) 313+ \( -name 'hccl_aiv_*_op_910_95.o' -o -name 'hccl_aiv_*_op_910_96.o' \) -print0)
308 log "Info: Total AIV kernels embedded: ${AIV_EMBED_COUNT}" 314 log "Info: Total AIV kernels embedded: ${AIV_EMBED_COUNT}"
309 315
310 316
@@ -974,6 +980,7 @@ CUSTOM_OPTION="${CUSTOM_OPTION} -DCANN_3RD_LIB_PATH=${CANN_3RD_LIB_PATH}"
974CUSTOM_OPTION="${CUSTOM_OPTION} -DCMAKE_BUILD_TYPE=${BUILD_TYPE}"980CUSTOM_OPTION="${CUSTOM_OPTION} -DCMAKE_BUILD_TYPE=${BUILD_TYPE}"
975CUSTOM_OPTION="${CUSTOM_OPTION} -DENABLE_BUILD_DEVICE=${ENABLE_BUILD_DEVICE}"981CUSTOM_OPTION="${CUSTOM_OPTION} -DENABLE_BUILD_DEVICE=${ENABLE_BUILD_DEVICE}"
976CUSTOM_OPTION="${CUSTOM_OPTION} -DENABLE_BUILD_AARCH=${ENABLE_BUILD_AARCH}"982CUSTOM_OPTION="${CUSTOM_OPTION} -DENABLE_BUILD_AARCH=${ENABLE_BUILD_AARCH}"
983+CUSTOM_OPTION="${CUSTOM_OPTION} -DHCCL_ENABLE_910_96=${ENABLE_910_96}"
977CUSTOM_OPTION="${CUSTOM_OPTION} -DCUSTOM_SIGN_SCRIPT=${CUSTOM_SIGN_SCRIPT}"984CUSTOM_OPTION="${CUSTOM_OPTION} -DCUSTOM_SIGN_SCRIPT=${CUSTOM_SIGN_SCRIPT}"
978CUSTOM_OPTION="${CUSTOM_OPTION} -DENABLE_SIGN=${ENABLE_SIGN}"985CUSTOM_OPTION="${CUSTOM_OPTION} -DENABLE_SIGN=${ENABLE_SIGN}"
979CUSTOM_OPTION="${CUSTOM_OPTION} -DVERSION_INFO=${VERSION_INFO}"986CUSTOM_OPTION="${CUSTOM_OPTION} -DVERSION_INFO=${VERSION_INFO}"
Mscripts/package/module/ascend/HcclLib.xml+50-0
@@ -82,6 +82,56 @@
82 <file value="hccl_a2a_superkernel_mesh_1d_uint32_v2.o" optional="true"/>82 <file value="hccl_a2a_superkernel_mesh_1d_uint32_v2.o" optional="true"/>
83 <file value="hccl_a2a_superkernel_mesh_1d_int64_v2.o" optional="true"/>83 <file value="hccl_a2a_superkernel_mesh_1d_int64_v2.o" optional="true"/>
84 <file value="hccl_a2a_superkernel_mesh_1d_uint64_v2.o" optional="true"/>84 <file value="hccl_a2a_superkernel_mesh_1d_uint64_v2.o" optional="true"/>
85+ <file value="hccl_aiv_reduce_scatter_op_910_96.o" optional="true"/>
86+ <file value="hccl_aiv_all_gather_op_910_96.o" optional="true"/>
87+ <file value="hccl_aiv_scatter_op_910_96.o" optional="true"/>
88+ <file value="hccl_aiv_all_reduce_op_910_96.o" optional="true"/>
89+ <file value="hccl_aiv_broadcast_op_910_96.o" optional="true"/>
90+ <file value="hccl_aiv_reduce_op_910_96.o" optional="true"/>
91+ <file value="hccl_aiv_all_to_all_v_op_910_96.o" optional="true"/>
92+ <file value="hccl_aiv_all_to_all_op_910_96.o" optional="true"/>
93+ <file value="hccl_aiv_send_op_910_96.o" optional="true"/>
94+ <file value="hccl_aiv_recv_op_910_96.o" optional="true"/>
95+ <file value="hccl_ar_superkernel_mesh_1d_oneshot_int8_v2_910_96.o" optional="true"/>
96+ <file value="hccl_ar_superkernel_mesh_1d_oneshot_int16_v2_910_96.o" optional="true"/>
97+ <file value="hccl_ar_superkernel_mesh_1d_oneshot_int32_v2_910_96.o" optional="true"/>
98+ <file value="hccl_ar_superkernel_mesh_1d_oneshot_float16_v2_910_96.o" optional="true"/>
99+ <file value="hccl_ar_superkernel_mesh_1d_oneshot_float32_v2_910_96.o" optional="true"/>
100+ <file value="hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2_910_96.o" optional="true"/>
101+ <file value="hccl_ar_superkernel_mesh_1d_twoshot_int8_v2_910_96.o" optional="true"/>
102+ <file value="hccl_ar_superkernel_mesh_1d_twoshot_int16_v2_910_96.o" optional="true"/>
103+ <file value="hccl_ar_superkernel_mesh_1d_twoshot_int32_v2_910_96.o" optional="true"/>
104+ <file value="hccl_ar_superkernel_mesh_1d_twoshot_float16_v2_910_96.o" optional="true"/>
105+ <file value="hccl_ar_superkernel_mesh_1d_twoshot_float32_v2_910_96.o" optional="true"/>
106+ <file value="hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2_910_96.o" optional="true"/>
107+ <file value="hccl_rs_superkernel_mesh_1d_int8_v2_910_96.o" optional="true"/>
108+ <file value="hccl_rs_superkernel_mesh_1d_int16_v2_910_96.o" optional="true"/>
109+ <file value="hccl_rs_superkernel_mesh_1d_int32_v2_910_96.o" optional="true"/>
110+ <file value="hccl_rs_superkernel_mesh_1d_float16_v2_910_96.o" optional="true"/>
111+ <file value="hccl_rs_superkernel_mesh_1d_float32_v2_910_96.o" optional="true"/>
112+ <file value="hccl_rs_superkernel_mesh_1d_bfloat16_v2_910_96.o" optional="true"/>
113+ <file value="hccl_ag_superkernel_mesh_1d_int8_v2_910_96.o" optional="true"/>
114+ <file value="hccl_ag_superkernel_mesh_1d_int16_v2_910_96.o" optional="true"/>
115+ <file value="hccl_ag_superkernel_mesh_1d_int32_v2_910_96.o" optional="true"/>
116+ <file value="hccl_ag_superkernel_mesh_1d_float16_v2_910_96.o" optional="true"/>
117+ <file value="hccl_ag_superkernel_mesh_1d_float32_v2_910_96.o" optional="true"/>
118+ <file value="hccl_ag_superkernel_mesh_1d_bfloat16_v2_910_96.o" optional="true"/>
119+ <file value="hccl_ag_superkernel_mesh_1d_uint8_v2_910_96.o" optional="true"/>
120+ <file value="hccl_ag_superkernel_mesh_1d_uint16_v2_910_96.o" optional="true"/>
121+ <file value="hccl_ag_superkernel_mesh_1d_uint32_v2_910_96.o" optional="true"/>
122+ <file value="hccl_ag_superkernel_mesh_1d_int64_v2_910_96.o" optional="true"/>
123+ <file value="hccl_ag_superkernel_mesh_1d_uint64_v2_910_96.o" optional="true"/>
124+ <file value="hccl_a2a_superkernel_mesh_1d_int8_v2_910_96.o" optional="true"/>
125+ <file value="hccl_a2a_superkernel_mesh_1d_int16_v2_910_96.o" optional="true"/>
126+ <file value="hccl_a2a_superkernel_mesh_1d_int32_v2_910_96.o" optional="true"/>
127+ <file value="hccl_a2a_superkernel_mesh_1d_float16_v2_910_96.o" optional="true"/>
128+ <file value="hccl_a2a_superkernel_mesh_1d_float32_v2_910_96.o" optional="true"/>
129+ <file value="hccl_a2a_superkernel_mesh_1d_bfloat16_v2_910_96.o" optional="true"/>
130+ <file value="hccl_a2a_superkernel_mesh_1d_uint8_v2_910_96.o" optional="true"/>
131+ <file value="hccl_a2a_superkernel_mesh_1d_uint16_v2_910_96.o" optional="true"/>
132+ <file value="hccl_a2a_superkernel_mesh_1d_uint32_v2_910_96.o" optional="true"/>
133+ <file value="hccl_a2a_superkernel_mesh_1d_int64_v2_910_96.o" optional="true"/>
134+ <file value="hccl_a2a_superkernel_mesh_1d_uint64_v2_910_96.o" optional="true"/>
85 </file_info>135 </file_info>
86 <file_info value="device" copy_type="delivery" src_path="build" dst_path="opp/built-in/op_impl/aicpu/kernel" install_path="opp/built-in/op_impl/aicpu/kernel" install_type="all" install_mod="440" optional="true">136 <file_info value="device" copy_type="delivery" src_path="build" dst_path="opp/built-in/op_impl/aicpu/kernel" install_path="opp/built-in/op_impl/aicpu/kernel" install_type="all" install_mod="440" optional="true">
87 <file value="aicpu_hccl.tar.gz"/>137 <file value="aicpu_hccl.tar.gz"/>
Msrc/CMakeLists.txt+50-0
@@ -316,6 +316,56 @@ set(AIV_TARGETS
316 hccl_a2a_superkernel_mesh_1d_uint32_v2316 hccl_a2a_superkernel_mesh_1d_uint32_v2
317 hccl_a2a_superkernel_mesh_1d_int64_v2317 hccl_a2a_superkernel_mesh_1d_int64_v2
318 hccl_a2a_superkernel_mesh_1d_uint64_v2318 hccl_a2a_superkernel_mesh_1d_uint64_v2
319+ hccl_aiv_all_gather_op_910_96
320+ hccl_aiv_all_reduce_op_910_96
321+ hccl_aiv_all_to_all_v_op_910_96
322+ hccl_aiv_all_to_all_op_910_96
323+ hccl_aiv_broadcast_op_910_96
324+ hccl_aiv_reduce_op_910_96
325+ hccl_aiv_reduce_scatter_op_910_96
326+ hccl_aiv_scatter_op_910_96
327+ hccl_aiv_send_op_910_96
328+ hccl_aiv_recv_op_910_96
329+ hccl_ar_superkernel_mesh_1d_oneshot_int8_v2_910_96
330+ hccl_ar_superkernel_mesh_1d_oneshot_int16_v2_910_96
331+ hccl_ar_superkernel_mesh_1d_oneshot_int32_v2_910_96
332+ hccl_ar_superkernel_mesh_1d_oneshot_float16_v2_910_96
333+ hccl_ar_superkernel_mesh_1d_oneshot_float32_v2_910_96
334+ hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2_910_96
335+ hccl_ar_superkernel_mesh_1d_twoshot_int8_v2_910_96
336+ hccl_ar_superkernel_mesh_1d_twoshot_int16_v2_910_96
337+ hccl_ar_superkernel_mesh_1d_twoshot_int32_v2_910_96
338+ hccl_ar_superkernel_mesh_1d_twoshot_float16_v2_910_96
339+ hccl_ar_superkernel_mesh_1d_twoshot_float32_v2_910_96
340+ hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2_910_96
341+ hccl_rs_superkernel_mesh_1d_int8_v2_910_96
342+ hccl_rs_superkernel_mesh_1d_int16_v2_910_96
343+ hccl_rs_superkernel_mesh_1d_int32_v2_910_96
344+ hccl_rs_superkernel_mesh_1d_float16_v2_910_96
345+ hccl_rs_superkernel_mesh_1d_float32_v2_910_96
346+ hccl_rs_superkernel_mesh_1d_bfloat16_v2_910_96
347+ hccl_ag_superkernel_mesh_1d_int8_v2_910_96
348+ hccl_ag_superkernel_mesh_1d_int16_v2_910_96
349+ hccl_ag_superkernel_mesh_1d_int32_v2_910_96
350+ hccl_ag_superkernel_mesh_1d_float16_v2_910_96
351+ hccl_ag_superkernel_mesh_1d_float32_v2_910_96
352+ hccl_ag_superkernel_mesh_1d_bfloat16_v2_910_96
353+ hccl_ag_superkernel_mesh_1d_uint8_v2_910_96
354+ hccl_ag_superkernel_mesh_1d_uint16_v2_910_96
355+ hccl_ag_superkernel_mesh_1d_uint32_v2_910_96
356+ hccl_ag_superkernel_mesh_1d_int64_v2_910_96
357+ hccl_ag_superkernel_mesh_1d_uint64_v2_910_96
358+ hccl_a2a_superkernel_mesh_1d_int8_v2_910_96
359+ hccl_a2a_superkernel_mesh_1d_int16_v2_910_96
360+ hccl_a2a_superkernel_mesh_1d_int32_v2_910_96
361+ hccl_a2a_superkernel_mesh_1d_float16_v2_910_96
362+ hccl_a2a_superkernel_mesh_1d_float32_v2_910_96
363+ hccl_a2a_superkernel_mesh_1d_bfloat16_v2_910_96
364+ hccl_a2a_superkernel_mesh_1d_uint8_v2_910_96
365+ hccl_a2a_superkernel_mesh_1d_uint16_v2_910_96
366+ hccl_a2a_superkernel_mesh_1d_uint32_v2_910_96
367+ hccl_a2a_superkernel_mesh_1d_int64_v2_910_96
368+ hccl_a2a_superkernel_mesh_1d_uint64_v2_910_96
319)369)
320 370 
321add_custom_target(aiv_all_targets)371add_custom_target(aiv_all_targets)
Msrc/hccl.cmake+4-0
@@ -61,6 +61,10 @@ if(HCCL_CANN_COMPAT_850)
61 target_compile_definitions(hccl PRIVATE HCCL_CANN_COMPAT_850)61 target_compile_definitions(hccl PRIVATE HCCL_CANN_COMPAT_850)
62endif()62endif()
63 63 
64+if(HCCL_ENABLE_910_96)
65+ target_compile_definitions(hccl PRIVATE HCCL_ENABLE_910_96)
66+endif()
67+ 
64target_compile_options(hccl PRIVATE68target_compile_options(hccl PRIVATE
65 -Werror69 -Werror
66 -Wno-error=float-equal70 -Wno-error=float-equal
Msrc/ops/all_gather/template/aiv/kernel/CMakeLists.txt+335-0
@@ -323,4 +323,339 @@ install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint64_v2.
323 COMPONENT hccl323 COMPONENT hccl
324)324)
325 325 
326+if(HCCL_ENABLE_910_96)
327+add_custom_target(hccl_aiv_all_gather_op_910_96 ALL
328+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_96.o)
329+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_96.o
330+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_all_gather_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec
331+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000
332+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
333+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
334+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
335+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
336+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
337+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
338+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
339+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
340+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
341+ COMMAND ${CCE_DIR}/ld.lld -m aicorelinux -Ttext=0 ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_96_tmp.o -static -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_96.o
342+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
343+)
344+ 
345+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_96.o
346+ DESTINATION ${INSTALL_LIBRARY_DIR}
347+ ${INSTALL_OPTIONAL}
348+ COMPONENT hccl
349+)
350+ 
351+add_custom_target(hccl_ag_superkernel_mesh_1d_int8_v2_910_96 ALL
352+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int8_v2_910_96.o)
353+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int8_v2_910_96.o
354+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int8_v2_910_96.o
355+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
356+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
357+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
358+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
359+ -DHCCL_DTYPE_INT8 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int8
360+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
361+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
362+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
363+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
364+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
365+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
366+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
367+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
368+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
369+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
370+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
371+)
372+ 
373+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int8_v2_910_96.o
374+ DESTINATION ${INSTALL_LIBRARY_DIR}
375+ ${INSTALL_OPTIONAL}
376+ COMPONENT hccl
377+)
378+ 
379+add_custom_target(hccl_ag_superkernel_mesh_1d_int16_v2_910_96 ALL
380+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int16_v2_910_96.o)
381+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int16_v2_910_96.o
382+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int16_v2_910_96.o
383+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
384+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
385+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
386+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
387+ -DHCCL_DTYPE_INT16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int16
388+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
389+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
390+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
391+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
392+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
393+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
394+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
395+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
396+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
397+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
398+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
399+)
400+ 
401+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int16_v2_910_96.o
402+ DESTINATION ${INSTALL_LIBRARY_DIR}
403+ ${INSTALL_OPTIONAL}
404+ COMPONENT hccl
405+)
406+ 
407+add_custom_target(hccl_ag_superkernel_mesh_1d_int32_v2_910_96 ALL
408+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int32_v2_910_96.o)
409+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int32_v2_910_96.o
410+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int32_v2_910_96.o
411+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
412+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
413+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
414+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
415+ -DHCCL_DTYPE_INT32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int32
416+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
417+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
418+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
419+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
420+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
421+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
422+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
423+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
424+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
425+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
426+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
427+)
428+ 
429+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int32_v2_910_96.o
430+ DESTINATION ${INSTALL_LIBRARY_DIR}
431+ ${INSTALL_OPTIONAL}
432+ COMPONENT hccl
433+)
434+ 
435+add_custom_target(hccl_ag_superkernel_mesh_1d_float16_v2_910_96 ALL
436+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float16_v2_910_96.o)
437+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float16_v2_910_96.o
438+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float16_v2_910_96.o
439+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
440+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
441+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
442+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
443+ -DHCCL_DTYPE_FP16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_float16
444+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
445+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
446+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
447+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
448+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
449+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
450+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
451+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
452+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
453+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
454+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
455+)
456+ 
457+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float16_v2_910_96.o
458+ DESTINATION ${INSTALL_LIBRARY_DIR}
459+ ${INSTALL_OPTIONAL}
460+ COMPONENT hccl
461+)
462+ 
463+add_custom_target(hccl_ag_superkernel_mesh_1d_float32_v2_910_96 ALL
464+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float32_v2_910_96.o)
465+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float32_v2_910_96.o
466+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float32_v2_910_96.o
467+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
468+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
469+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
470+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
471+ -DHCCL_DTYPE_FP32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_float32
472+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
473+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
474+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
475+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
476+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
477+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
478+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
479+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
480+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
481+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
482+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
483+)
484+ 
485+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float32_v2_910_96.o
486+ DESTINATION ${INSTALL_LIBRARY_DIR}
487+ ${INSTALL_OPTIONAL}
488+ COMPONENT hccl
489+)
490+ 
491+add_custom_target(hccl_ag_superkernel_mesh_1d_bfloat16_v2_910_96 ALL
492+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_bfloat16_v2_910_96.o)
493+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_bfloat16_v2_910_96.o
494+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_bfloat16_v2_910_96.o
495+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
496+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
497+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
498+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
499+ -DHCCL_DTYPE_BFP16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_bfloat16
500+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
501+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
502+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
503+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
504+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
505+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
506+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
507+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
508+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
509+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
510+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
511+)
512+ 
513+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_bfloat16_v2_910_96.o
514+ DESTINATION ${INSTALL_LIBRARY_DIR}
515+ ${INSTALL_OPTIONAL}
516+ COMPONENT hccl
517+)
518+ 
519+add_custom_target(hccl_ag_superkernel_mesh_1d_uint8_v2_910_96 ALL
520+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint8_v2_910_96.o)
521+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint8_v2_910_96.o
522+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint8_v2_910_96.o
523+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
524+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
525+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
526+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
527+ -DHCCL_DTYPE_UINT8 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint8
528+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
529+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
530+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
531+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
532+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
533+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
534+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
535+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
536+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
537+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
538+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
539+)
540+ 
541+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint8_v2_910_96.o
542+ DESTINATION ${INSTALL_LIBRARY_DIR}
543+ ${INSTALL_OPTIONAL}
544+ COMPONENT hccl
545+)
546+ 
547+add_custom_target(hccl_ag_superkernel_mesh_1d_uint16_v2_910_96 ALL
548+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint16_v2_910_96.o)
549+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint16_v2_910_96.o
550+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint16_v2_910_96.o
551+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
552+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
553+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
554+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
555+ -DHCCL_DTYPE_UINT16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint16
556+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
557+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
558+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
559+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
560+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
561+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
562+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
563+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
564+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
565+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
566+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
567+)
568+ 
569+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint16_v2_910_96.o
570+ DESTINATION ${INSTALL_LIBRARY_DIR}
571+ ${INSTALL_OPTIONAL}
572+ COMPONENT hccl
573+)
574+ 
575+add_custom_target(hccl_ag_superkernel_mesh_1d_uint32_v2_910_96 ALL
576+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint32_v2_910_96.o)
577+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint32_v2_910_96.o
578+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint32_v2_910_96.o
579+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
580+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
581+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
582+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
583+ -DHCCL_DTYPE_UINT32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint32
584+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
585+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
586+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
587+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
588+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
589+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
590+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
591+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
592+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
593+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
594+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
595+)
596+ 
597+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint32_v2_910_96.o
598+ DESTINATION ${INSTALL_LIBRARY_DIR}
599+ ${INSTALL_OPTIONAL}
600+ COMPONENT hccl
601+)
602+ 
603+add_custom_target(hccl_ag_superkernel_mesh_1d_int64_v2_910_96 ALL
604+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int64_v2_910_96.o)
605+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int64_v2_910_96.o
606+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int64_v2_910_96.o
607+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
608+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
609+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
610+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
611+ -DHCCL_DTYPE_INT64 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int64
612+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
613+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
614+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
615+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
616+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
617+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
618+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
619+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
620+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
621+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
622+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
623+)
624+ 
625+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int64_v2_910_96.o
626+ DESTINATION ${INSTALL_LIBRARY_DIR}
627+ ${INSTALL_OPTIONAL}
628+ COMPONENT hccl
629+)
630+ 
631+add_custom_target(hccl_ag_superkernel_mesh_1d_uint64_v2_910_96 ALL
632+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint64_v2_910_96.o)
633+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint64_v2_910_96.o
634+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint64_v2_910_96.o
635+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
636+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
637+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
638+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
639+ -DHCCL_DTYPE_UINT64 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint64
640+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
641+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
642+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
643+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
644+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
645+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
646+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
647+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
648+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
649+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
650+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
651+)
652+ 
653+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint64_v2_910_96.o
654+ DESTINATION ${INSTALL_LIBRARY_DIR}
655+ ${INSTALL_OPTIONAL}
656+ COMPONENT hccl
657+)
658+ 
659+endif()
660+ 
326endif()661endif()
Msrc/ops/all_reduce/all_reduce_op.cc+4-0
@@ -25,12 +25,14 @@ HcclResult HcclAllReduce(void *sendBuf, void *recvBuf, uint64_t count, HcclDataT
25{25{
26 HCCL_INFO("Start to run execute HcclAllReduce");26 HCCL_INFO("Start to run execute HcclAllReduce");
27 if (GetHcommVersion() < CANN_VERSION(9, 0, 0)) { // compat handle27 if (GetHcommVersion() < CANN_VERSION(9, 0, 0)) { // compat handle
28+ HCCL_INFO("werwer0001");
28 return HcclAllReduceInner(sendBuf, recvBuf, count, dataType, op, comm, stream);29 return HcclAllReduceInner(sendBuf, recvBuf, count, dataType, op, comm, stream);
29 }30 }
30 31 
31 bool isOutPlace = false;32 bool isOutPlace = false;
32 CHK_RET(IsOutPlaceDevice(isOutPlace));33 CHK_RET(IsOutPlaceDevice(isOutPlace));
33 if (!isOutPlace) {34 if (!isOutPlace) {
35+ HCCL_INFO("werwer0002");
34 return HcclAllReduceInner(sendBuf, recvBuf, count, dataType, op, comm, stream);36 return HcclAllReduceInner(sendBuf, recvBuf, count, dataType, op, comm, stream);
35 }37 }
36 CHK_PRT_RET(count == 0, HCCL_WARNING("input count is 0, return all reduce success"), HCCL_SUCCESS);38 CHK_PRT_RET(count == 0, HCCL_WARNING("input count is 0, return all reduce success"), HCCL_SUCCESS);
@@ -185,6 +187,7 @@ HcclResult AllReduceOutPlaceCommon(void *sendBuf, void *recvBuf, uint64_t count,
185 // 9.0.0 ccu模式走老流程187 // 9.0.0 ccu模式走老流程
186 if (opMode == OpMode::OPBASE && GetHcommVersion() == CANN_VERSION(9, 0, 0) &&188 if (opMode == OpMode::OPBASE && GetHcommVersion() == CANN_VERSION(9, 0, 0) &&
187 param.engine == CommEngine::COMM_ENGINE_CCU) {189 param.engine == CommEngine::COMM_ENGINE_CCU) {
190+ HCCL_INFO("werwer0003");
188 return HcclAllReduceInner(sendBuf, recvBuf, count, dataType, op, comm, stream);191 return HcclAllReduceInner(sendBuf, recvBuf, count, dataType, op, comm, stream);
189 }192 }
190 193 
@@ -205,6 +208,7 @@ HcclResult AllReduceOutPlaceCommon(void *sendBuf, void *recvBuf, uint64_t count,
205 std::unique_ptr<TopoInfoWithNetLayerDetails> topoInfo = std::make_unique<TopoInfoWithNetLayerDetails>();208 std::unique_ptr<TopoInfoWithNetLayerDetails> topoInfo = std::make_unique<TopoInfoWithNetLayerDetails>();
206 CHK_RET(Selector(comm, param, topoInfo, algName));209 CHK_RET(Selector(comm, param, topoInfo, algName));
207 if (ShouldUseInnerOp(param.opExecuteConfig) && param.opMode == OpMode::OPBASE) {210 if (ShouldUseInnerOp(param.opExecuteConfig) && param.opMode == OpMode::OPBASE) {
211+ HCCL_INFO("werwer0004");
208 return HcclAllReduceInner(sendBuf, recvBuf, count, dataType, op, comm, stream);212 return HcclAllReduceInner(sendBuf, recvBuf, count, dataType, op, comm, stream);
209 }213 }
210 // 单卡校验214 // 单卡校验
Msrc/ops/all_reduce/template/aiv/kernel/CMakeLists.txt+363-0
@@ -350,4 +350,367 @@ install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_bf
350 COMPONENT hccl350 COMPONENT hccl
351)351)
352 352 
353+if(HCCL_ENABLE_910_96)
354+add_custom_target(hccl_aiv_all_reduce_op_910_96 ALL
355+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_96.o)
356+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_96.o
357+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_all_reduce_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec
358+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000
359+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
360+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
361+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
362+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
363+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
364+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
365+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
366+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
367+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
368+ COMMAND ${CCE_DIR}/ld.lld -m aicorelinux -Ttext=0 ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_96_tmp.o -static -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_96.o
369+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
370+)
371+ 
372+add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_int8_v2_910_96 ALL
373+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int8_v2_910_96.o)
374+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int8_v2_910_96.o
375+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int8_v2_910_96.o
376+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
377+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
378+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
379+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
380+ -DHCCL_DTYPE_INT8 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int8
381+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
382+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
383+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
384+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
385+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
386+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
387+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
388+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
389+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
390+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
391+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
392+)
393+ 
394+add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_int16_v2_910_96 ALL
395+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int16_v2_910_96.o)
396+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int16_v2_910_96.o
397+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int16_v2_910_96.o
398+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
399+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
400+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
401+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
402+ -DHCCL_DTYPE_INT16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int16
403+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
404+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
405+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
406+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
407+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
408+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
409+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
410+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
411+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
412+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
413+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
414+)
415+ 
416+add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_int32_v2_910_96 ALL
417+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int32_v2_910_96.o)
418+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int32_v2_910_96.o
419+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int32_v2_910_96.o
420+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
421+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
422+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
423+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
424+ -DHCCL_DTYPE_INT32 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int32
425+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
426+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
427+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
428+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
429+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
430+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
431+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
432+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
433+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
434+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
435+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
436+)
437+ 
438+add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_float16_v2_910_96 ALL
439+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float16_v2_910_96.o)
440+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float16_v2_910_96.o
441+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float16_v2_910_96.o
442+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
443+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
444+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
445+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
446+ -DHCCL_DTYPE_FP16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_float16
447+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
448+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
449+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
450+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
451+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
452+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
453+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
454+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
455+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
456+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
457+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
458+)
459+ 
460+add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_float32_v2_910_96 ALL
461+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float32_v2_910_96.o)
462+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float32_v2_910_96.o
463+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float32_v2_910_96.o
464+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
465+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
466+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
467+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
468+ -DHCCL_DTYPE_FP32 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_float32
469+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
470+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
471+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
472+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
473+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
474+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
475+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
476+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
477+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
478+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
479+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
480+)
481+ 
482+add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2_910_96 ALL
483+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2_910_96.o)
484+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2_910_96.o
485+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2_910_96.o
486+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
487+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
488+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
489+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
490+ -DHCCL_DTYPE_BFP16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_bfloat16
491+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
492+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
493+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
494+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
495+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
496+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
497+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
498+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
499+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
500+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
501+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
502+)
503+ 
504+add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_int8_v2_910_96 ALL
505+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int8_v2_910_96.o)
506+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int8_v2_910_96.o
507+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int8_v2_910_96.o
508+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
509+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
510+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
511+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
512+ -DHCCL_DTYPE_INT8 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int8
513+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
514+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
515+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
516+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
517+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
518+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
519+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
520+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
521+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
522+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
523+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
524+)
525+ 
526+add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_int16_v2_910_96 ALL
527+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int16_v2_910_96.o)
528+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int16_v2_910_96.o
529+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int16_v2_910_96.o
530+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
531+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
532+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
533+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
534+ -DHCCL_DTYPE_INT16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int16
535+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
536+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
537+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
538+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
539+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
540+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
541+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
542+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
543+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
544+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
545+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
546+)
547+ 
548+add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_int32_v2_910_96 ALL
549+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int32_v2_910_96.o)
550+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int32_v2_910_96.o
551+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int32_v2_910_96.o
552+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
553+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
554+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
555+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
556+ -DHCCL_DTYPE_INT32 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int32
557+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
558+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
559+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
560+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
561+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
562+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
563+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
564+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
565+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
566+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
567+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
568+)
569+ 
570+add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_float16_v2_910_96 ALL
571+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float16_v2_910_96.o)
572+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float16_v2_910_96.o
573+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float16_v2_910_96.o
574+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
575+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
576+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
577+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
578+ -DHCCL_DTYPE_FP16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_float16
579+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
580+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
581+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
582+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
583+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
584+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
585+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
586+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
587+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
588+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
589+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
590+)
591+ 
592+add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_float32_v2_910_96 ALL
593+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float32_v2_910_96.o)
594+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float32_v2_910_96.o
595+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float32_v2_910_96.o
596+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
597+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
598+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
599+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
600+ -DHCCL_DTYPE_FP32 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_float32
601+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
602+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
603+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
604+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
605+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
606+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
607+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
608+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
609+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
610+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
611+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
612+)
613+ 
614+add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2_910_96 ALL
615+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2_910_96.o)
616+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2_910_96.o
617+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2_910_96.o
618+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
619+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
620+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
621+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
622+ -DHCCL_DTYPE_BFP16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_bfloat16
623+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
624+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
625+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
626+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
627+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
628+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
629+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
630+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
631+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
632+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
633+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
634+)
635+ 
636+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_96.o
637+ DESTINATION ${INSTALL_LIBRARY_DIR}
638+ ${INSTALL_OPTIONAL}
639+ COMPONENT hccl
640+)
641+ 
642+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int8_v2_910_96.o
643+ DESTINATION ${INSTALL_LIBRARY_DIR}
644+ ${INSTALL_OPTIONAL}
645+ COMPONENT hccl
646+)
647+ 
648+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int16_v2_910_96.o
649+ DESTINATION ${INSTALL_LIBRARY_DIR}
650+ ${INSTALL_OPTIONAL}
651+ COMPONENT hccl
652+)
653+ 
654+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int32_v2_910_96.o
655+ DESTINATION ${INSTALL_LIBRARY_DIR}
656+ ${INSTALL_OPTIONAL}
657+ COMPONENT hccl
658+)
659+ 
660+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float16_v2_910_96.o
661+ DESTINATION ${INSTALL_LIBRARY_DIR}
662+ ${INSTALL_OPTIONAL}
663+ COMPONENT hccl
664+)
665+ 
666+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float32_v2_910_96.o
667+ DESTINATION ${INSTALL_LIBRARY_DIR}
668+ ${INSTALL_OPTIONAL}
669+ COMPONENT hccl
670+)
671+ 
672+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2_910_96.o
673+ DESTINATION ${INSTALL_LIBRARY_DIR}
674+ ${INSTALL_OPTIONAL}
675+ COMPONENT hccl
676+)
677+ 
678+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int8_v2_910_96.o
679+ DESTINATION ${INSTALL_LIBRARY_DIR}
680+ ${INSTALL_OPTIONAL}
681+ COMPONENT hccl
682+)
683+ 
684+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int16_v2_910_96.o
685+ DESTINATION ${INSTALL_LIBRARY_DIR}
686+ ${INSTALL_OPTIONAL}
687+ COMPONENT hccl
688+)
689+ 
690+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int32_v2_910_96.o
691+ DESTINATION ${INSTALL_LIBRARY_DIR}
692+ ${INSTALL_OPTIONAL}
693+ COMPONENT hccl
694+)
695+ 
696+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float16_v2_910_96.o
697+ DESTINATION ${INSTALL_LIBRARY_DIR}
698+ ${INSTALL_OPTIONAL}
699+ COMPONENT hccl
700+)
701+ 
702+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float32_v2_910_96.o
703+ DESTINATION ${INSTALL_LIBRARY_DIR}
704+ ${INSTALL_OPTIONAL}
705+ COMPONENT hccl
706+)
707+ 
708+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2_910_96.o
709+ DESTINATION ${INSTALL_LIBRARY_DIR}
710+ ${INSTALL_OPTIONAL}
711+ COMPONENT hccl
712+)
713+ 
714+endif()
715+ 
353endif()716endif()
Msrc/ops/all_to_all_v/template/aiv/kernel/CMakeLists.txt+359-0
@@ -346,4 +346,363 @@ install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint64_v2
346 COMPONENT hccl346 COMPONENT hccl
347)347)
348 348 
349+if(HCCL_ENABLE_910_96)
350+add_custom_target(hccl_aiv_all_to_all_v_op_910_96 ALL
351+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_96.o)
352+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_96.o
353+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_all_to_all_v_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec
354+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000
355+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
356+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
357+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
358+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
359+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
360+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
361+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
362+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
363+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
364+ COMMAND ${CCE_DIR}/ld.lld -m aicorelinux -Ttext=0 ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_96_tmp.o -static -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_96.o
365+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
366+)
367+ 
368+add_custom_target(hccl_aiv_all_to_all_op_910_96 ALL
369+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_96.o)
370+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_96.o
371+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_all_to_all_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec
372+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000
373+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
374+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
375+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
376+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
377+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
378+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
379+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
380+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
381+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
382+ COMMAND ${CCE_DIR}/ld.lld -m aicorelinux -Ttext=0 ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_96_tmp.o -static -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_96.o
383+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
384+)
385+ 
386+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_96.o
387+ DESTINATION ${INSTALL_LIBRARY_DIR}
388+ ${INSTALL_OPTIONAL}
389+ COMPONENT hccl
390+)
391+ 
392+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_96.o
393+ DESTINATION ${INSTALL_LIBRARY_DIR}
394+ ${INSTALL_OPTIONAL}
395+ COMPONENT hccl
396+)
397+ 
398+add_custom_target(hccl_a2a_superkernel_mesh_1d_int8_v2_910_96 ALL
399+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int8_v2_910_96.o)
400+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int8_v2_910_96.o
401+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int8_v2_910_96.o
402+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
403+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
404+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
405+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
406+ -DHCCL_DTYPE_INT8 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int8
407+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
408+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
409+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
410+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
411+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
412+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
413+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
414+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
415+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
416+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
417+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
418+)
419+ 
420+add_custom_target(hccl_a2a_superkernel_mesh_1d_int16_v2_910_96 ALL
421+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int16_v2_910_96.o)
422+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int16_v2_910_96.o
423+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int16_v2_910_96.o
424+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
425+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
426+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
427+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
428+ -DHCCL_DTYPE_INT16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int16
429+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
430+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
431+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
432+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
433+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
434+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
435+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
436+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
437+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
438+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
439+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
440+)
441+ 
442+add_custom_target(hccl_a2a_superkernel_mesh_1d_int32_v2_910_96 ALL
443+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int32_v2_910_96.o)
444+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int32_v2_910_96.o
445+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int32_v2_910_96.o
446+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
447+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
448+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
449+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
450+ -DHCCL_DTYPE_INT32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int32
451+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
452+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
453+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
454+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
455+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
456+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
457+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
458+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
459+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
460+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
461+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
462+)
463+ 
464+add_custom_target(hccl_a2a_superkernel_mesh_1d_float16_v2_910_96 ALL
465+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float16_v2_910_96.o)
466+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float16_v2_910_96.o
467+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float16_v2_910_96.o
468+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
469+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
470+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
471+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
472+ -DHCCL_DTYPE_FP16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_float16
473+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
474+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
475+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
476+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
477+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
478+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
479+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
480+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
481+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
482+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
483+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
484+)
485+ 
486+add_custom_target(hccl_a2a_superkernel_mesh_1d_float32_v2_910_96 ALL
487+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float32_v2_910_96.o)
488+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float32_v2_910_96.o
489+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float32_v2_910_96.o
490+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
491+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
492+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
493+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
494+ -DHCCL_DTYPE_FP32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_float32
495+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
496+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
497+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
498+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
499+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
500+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
501+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
502+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
503+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
504+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
505+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
506+)
507+ 
508+add_custom_target(hccl_a2a_superkernel_mesh_1d_bfloat16_v2_910_96 ALL
509+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_bfloat16_v2_910_96.o)
510+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_bfloat16_v2_910_96.o
511+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_bfloat16_v2_910_96.o
512+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
513+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
514+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
515+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
516+ -DHCCL_DTYPE_BFP16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_bfloat16
517+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
518+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
519+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
520+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
521+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
522+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
523+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
524+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
525+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
526+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
527+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
528+)
529+ 
530+add_custom_target(hccl_a2a_superkernel_mesh_1d_uint8_v2_910_96 ALL
531+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint8_v2_910_96.o)
532+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint8_v2_910_96.o
533+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint8_v2_910_96.o
534+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
535+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
536+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
537+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
538+ -DHCCL_DTYPE_UINT8 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint8
539+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
540+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
541+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
542+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
543+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
544+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
545+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
546+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
547+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
548+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
549+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
550+)
551+ 
552+add_custom_target(hccl_a2a_superkernel_mesh_1d_uint16_v2_910_96 ALL
553+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint16_v2_910_96.o)
554+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint16_v2_910_96.o
555+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint16_v2_910_96.o
556+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
557+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
558+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
559+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
560+ -DHCCL_DTYPE_UINT16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint16
561+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
562+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
563+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
564+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
565+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
566+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
567+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
568+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
569+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
570+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
571+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
572+)
573+ 
574+add_custom_target(hccl_a2a_superkernel_mesh_1d_uint32_v2_910_96 ALL
575+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint32_v2_910_96.o)
576+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint32_v2_910_96.o
577+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint32_v2_910_96.o
578+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
579+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
580+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
581+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
582+ -DHCCL_DTYPE_UINT32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint32
583+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
584+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
585+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
586+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
587+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
588+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
589+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
590+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
591+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
592+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
593+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
594+)
595+ 
596+add_custom_target(hccl_a2a_superkernel_mesh_1d_int64_v2_910_96 ALL
597+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int64_v2_910_96.o)
598+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int64_v2_910_96.o
599+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int64_v2_910_96.o
600+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
601+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
602+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
603+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
604+ -DHCCL_DTYPE_INT64 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int64
605+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
606+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
607+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
608+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
609+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
610+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
611+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
612+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
613+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
614+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
615+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
616+)
617+ 
618+add_custom_target(hccl_a2a_superkernel_mesh_1d_uint64_v2_910_96 ALL
619+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint64_v2_910_96.o)
620+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint64_v2_910_96.o
621+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint64_v2_910_96.o
622+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
623+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
624+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
625+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
626+ -DHCCL_DTYPE_UINT64 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint64
627+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
628+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
629+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
630+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
631+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
632+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
633+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
634+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
635+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
636+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
637+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
638+)
639+ 
640+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int8_v2_910_96.o
641+ DESTINATION ${INSTALL_LIBRARY_DIR}
642+ ${INSTALL_OPTIONAL}
643+ COMPONENT hccl
644+)
645+ 
646+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int16_v2_910_96.o
647+ DESTINATION ${INSTALL_LIBRARY_DIR}
648+ ${INSTALL_OPTIONAL}
649+ COMPONENT hccl
650+)
651+ 
652+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int32_v2_910_96.o
653+ DESTINATION ${INSTALL_LIBRARY_DIR}
654+ ${INSTALL_OPTIONAL}
655+ COMPONENT hccl
656+)
657+ 
658+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float16_v2_910_96.o
659+ DESTINATION ${INSTALL_LIBRARY_DIR}
660+ ${INSTALL_OPTIONAL}
661+ COMPONENT hccl
662+)
663+ 
664+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float32_v2_910_96.o
665+ DESTINATION ${INSTALL_LIBRARY_DIR}
666+ ${INSTALL_OPTIONAL}
667+ COMPONENT hccl
668+)
669+ 
670+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_bfloat16_v2_910_96.o
671+ DESTINATION ${INSTALL_LIBRARY_DIR}
672+ ${INSTALL_OPTIONAL}
673+ COMPONENT hccl
674+)
675+ 
676+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint8_v2_910_96.o
677+ DESTINATION ${INSTALL_LIBRARY_DIR}
678+ ${INSTALL_OPTIONAL}
679+ COMPONENT hccl
680+)
681+ 
682+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint16_v2_910_96.o
683+ DESTINATION ${INSTALL_LIBRARY_DIR}
684+ ${INSTALL_OPTIONAL}
685+ COMPONENT hccl
686+)
687+ 
688+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint32_v2_910_96.o
689+ DESTINATION ${INSTALL_LIBRARY_DIR}
690+ ${INSTALL_OPTIONAL}
691+ COMPONENT hccl
692+)
693+ 
694+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int64_v2_910_96.o
695+ DESTINATION ${INSTALL_LIBRARY_DIR}
696+ ${INSTALL_OPTIONAL}
697+ COMPONENT hccl
698+)
699+ 
700+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint64_v2_910_96.o
701+ DESTINATION ${INSTALL_LIBRARY_DIR}
702+ ${INSTALL_OPTIONAL}
703+ COMPONENT hccl
704+)
705+ 
706+endif()
707+ 
349endif()708endif()
Msrc/ops/broadcast/template/aiv/kernel/CMakeLists.txt+27-0
@@ -25,4 +25,31 @@ install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_95.o
25 ${INSTALL_OPTIONAL}25 ${INSTALL_OPTIONAL}
26 COMPONENT hccl26 COMPONENT hccl
27)27)
28+ 
29+if(HCCL_ENABLE_910_96)
30+add_custom_target(hccl_aiv_broadcast_op_910_96 ALL
31+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_96.o)
32+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_96.o
33+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_broadcast_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec
34+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000
35+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
36+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
37+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
38+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
39+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
40+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
41+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
42+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
43+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
44+ COMMAND ${CCE_DIR}/ld.lld -m aicorelinux -Ttext=0 ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_96_tmp.o -static -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_96.o
45+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
46+)
47+ 
48+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_96.o
49+ DESTINATION ${INSTALL_LIBRARY_DIR}
50+ ${INSTALL_OPTIONAL}
51+ COMPONENT hccl
52+)
53+endif()
54+ 
28endif()55endif()
Msrc/ops/op_common/template/aiv/aiv_kernel_def.h+35-0
@@ -208,6 +208,41 @@ static std::map<HcclCMDType, std::pair<std::string, std::vector<AivKernelInfo>>>
208 {HcclCMDType::HCCL_CMD_SEND, {g_sendAivBinaryName, g_sendAivKernelInfoList}},208 {HcclCMDType::HCCL_CMD_SEND, {g_sendAivBinaryName, g_sendAivKernelInfoList}},
209 {HcclCMDType::HCCL_CMD_RECEIVE, {g_recvAivBinaryName, g_recvAivKernelInfoList}},209 {HcclCMDType::HCCL_CMD_RECEIVE, {g_recvAivBinaryName, g_recvAivKernelInfoList}},
210};210};
211+ 
212+// A6版本(dav-920r1-vec)对应的二进制名,kernel信息列表与A5版本一致
213+static std::string g_allreduceAivBinaryNameA6 = "hccl_aiv_all_reduce_op_910_96.o";
214+static std::string g_reduceScatterAivBinaryNameA6 = "hccl_aiv_reduce_scatter_op_910_96.o";
215+static std::string g_allgatherAivBinaryNameA6 = "hccl_aiv_all_gather_op_910_96.o";
216+static std::string g_broadcastAivBinaryNameA6 = "hccl_aiv_broadcast_op_910_96.o";
217+static std::string g_alltoallAivBinaryNameA6 = "hccl_aiv_all_to_all_op_910_96.o";
218+static std::string g_alltoallvAivBinaryNameA6 = "hccl_aiv_all_to_all_v_op_910_96.o";
219+static std::string g_scatterAivBinaryNameA6 = "hccl_aiv_scatter_op_910_96.o";
220+static std::string g_reduceAivBinaryNameA6 = "hccl_aiv_reduce_op_910_96.o";
221+static std::string g_sendAivBinaryNameA6 = "hccl_aiv_send_op_910_96.o";
222+static std::string g_recvAivBinaryNameA6 = "hccl_aiv_recv_op_910_96.o";
223+ 
224+static std::map<HcclCMDType, std::pair<std::string, std::vector<AivKernelInfo>>> g_aivKernelInfoMapA6 = {
225+ {HcclCMDType::HCCL_CMD_ALLREDUCE, {g_allreduceAivBinaryNameA6, g_allreduceAivKernelInfoList}},
226+ {HcclCMDType::HCCL_CMD_REDUCE_SCATTER, {g_reduceScatterAivBinaryNameA6, g_reduceScatterAivKernelInfoList}},
227+ {HcclCMDType::HCCL_CMD_ALLGATHER, {g_allgatherAivBinaryNameA6, g_allgatherAivKernelInfoList}},
228+ {HcclCMDType::HCCL_CMD_BROADCAST, {g_broadcastAivBinaryNameA6, g_broadcastAivKernelInfoList}},
229+ {HcclCMDType::HCCL_CMD_ALLTOALL, {g_alltoallAivBinaryNameA6, g_alltoallAivKernelInfoList}},
230+ {HcclCMDType::HCCL_CMD_ALLTOALLV, {g_alltoallvAivBinaryNameA6, g_alltoallvAivKernelInfoList}},
231+ {HcclCMDType::HCCL_CMD_SCATTER, {g_scatterAivBinaryNameA6, g_scatterAivKernelInfoList}},
232+ {HcclCMDType::HCCL_CMD_REDUCE, {g_reduceAivBinaryNameA6, g_reduceAivKernelInfoList}},
233+ {HcclCMDType::HCCL_CMD_SEND, {g_sendAivBinaryNameA6, g_sendAivKernelInfoList}},
234+ {HcclCMDType::HCCL_CMD_RECEIVE, {g_recvAivBinaryNameA6, g_recvAivKernelInfoList}},
235+};
236+ 
237+// 根据设备类型选择AIV kernel信息map:960芯片使用A6版本,其余使用A5版本
238+inline const std::map<HcclCMDType, std::pair<std::string, std::vector<AivKernelInfo>>>& GetAivKernelInfoMap(
239+ HcclDevType deviceType)
240+{
241+ if (deviceType == HcclDevType::DEV_TYPE_960) {
242+ return g_aivKernelInfoMapA6;
243+ }
244+ return g_aivKernelInfoMap;
245+}
211}246}
212 247 
213#endif // HCCL_AIV_UTILS_H248#endif // HCCL_AIV_UTILS_H
Msrc/ops/op_common/template/aiv/hccl_aiv_utils.cc+28-1
@@ -54,6 +54,18 @@ DECL_AIV_EMBED(hccl_aiv_reduce_scatter_op_910_95);
54DECL_AIV_EMBED(hccl_aiv_scatter_op_910_95);54DECL_AIV_EMBED(hccl_aiv_scatter_op_910_95);
55DECL_AIV_EMBED(hccl_aiv_send_op_910_95);55DECL_AIV_EMBED(hccl_aiv_send_op_910_95);
56DECL_AIV_EMBED(hccl_aiv_recv_op_910_95);56DECL_AIV_EMBED(hccl_aiv_recv_op_910_95);
57+#ifdef HCCL_ENABLE_910_96
58+DECL_AIV_EMBED(hccl_aiv_all_gather_op_910_96);
59+DECL_AIV_EMBED(hccl_aiv_all_reduce_op_910_96);
60+DECL_AIV_EMBED(hccl_aiv_all_to_all_op_910_96);
61+DECL_AIV_EMBED(hccl_aiv_all_to_all_v_op_910_96);
62+DECL_AIV_EMBED(hccl_aiv_broadcast_op_910_96);
63+DECL_AIV_EMBED(hccl_aiv_reduce_op_910_96);
64+DECL_AIV_EMBED(hccl_aiv_reduce_scatter_op_910_96);
65+DECL_AIV_EMBED(hccl_aiv_scatter_op_910_96);
66+DECL_AIV_EMBED(hccl_aiv_send_op_910_96);
67+DECL_AIV_EMBED(hccl_aiv_recv_op_910_96);
68+#endif // HCCL_ENABLE_910_96
57#undef DECL_AIV_EMBED69#undef DECL_AIV_EMBED
58}70}
59 71 
@@ -77,6 +89,18 @@ static const AivEmbedSymbol kAivEmbedTable[] = {
77 EMBED_ENTRY(hccl_aiv_scatter_op_910_95),89 EMBED_ENTRY(hccl_aiv_scatter_op_910_95),
78 EMBED_ENTRY(hccl_aiv_send_op_910_95),90 EMBED_ENTRY(hccl_aiv_send_op_910_95),
79 EMBED_ENTRY(hccl_aiv_recv_op_910_95),91 EMBED_ENTRY(hccl_aiv_recv_op_910_95),
92+#ifdef HCCL_ENABLE_910_96
93+ EMBED_ENTRY(hccl_aiv_all_gather_op_910_96),
94+ EMBED_ENTRY(hccl_aiv_all_reduce_op_910_96),
95+ EMBED_ENTRY(hccl_aiv_all_to_all_op_910_96),
96+ EMBED_ENTRY(hccl_aiv_all_to_all_v_op_910_96),
97+ EMBED_ENTRY(hccl_aiv_broadcast_op_910_96),
98+ EMBED_ENTRY(hccl_aiv_reduce_op_910_96),
99+ EMBED_ENTRY(hccl_aiv_reduce_scatter_op_910_96),
100+ EMBED_ENTRY(hccl_aiv_scatter_op_910_96),
101+ EMBED_ENTRY(hccl_aiv_send_op_910_96),
102+ EMBED_ENTRY(hccl_aiv_recv_op_910_96),
103+#endif // HCCL_ENABLE_910_96
80};104};
81#undef EMBED_ENTRY105#undef EMBED_ENTRY
82 106 
@@ -613,7 +637,10 @@ HcclResult RegisterKernel()
613 return HCCL_SUCCESS;637 return HCCL_SUCCESS;
614 }638 }
615 639 
616- for (const auto& item : g_aivKernelInfoMap) {640+ HcclDevType deviceType = HcclDevType::DEV_TYPE_COUNT;
641+ CHK_RET(HcclGetDeviceType(deviceType));
642+ 
643+ for (const auto& item : GetAivKernelInfoMap(deviceType)) {
617 const HcclCMDType cmdType = item.first;644 const HcclCMDType cmdType = item.first;
618 const std::string& aivBinaryName = item.second.first;645 const std::string& aivBinaryName = item.second.first;
619 const std::vector<AivKernelInfo>& aivKernelInfoList = item.second.second;646 const std::vector<AivKernelInfo>& aivKernelInfoList = item.second.second;
Msrc/ops/recv/template/aiv/kernel/CMakeLists.txt+27-0
@@ -25,4 +25,31 @@ install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_95.o
25 ${INSTALL_OPTIONAL}25 ${INSTALL_OPTIONAL}
26 COMPONENT hccl26 COMPONENT hccl
27)27)
28+ 
29+if(HCCL_ENABLE_910_96)
30+add_custom_target(hccl_aiv_recv_op_910_96 ALL
31+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_96.o)
32+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_96.o
33+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_recv_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec
34+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000
35+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
36+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
37+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
38+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
39+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
40+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
41+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
42+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
43+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
44+ COMMAND ${CCE_DIR}/ld.lld -m aicorelinux -Ttext=0 ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_96_tmp.o -static -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_96.o
45+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
46+)
47+ 
48+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_96.o
49+ DESTINATION ${INSTALL_LIBRARY_DIR}
50+ ${INSTALL_OPTIONAL}
51+ COMPONENT hccl
52+)
53+endif()
54+ 
28endif()55endif()
Msrc/ops/reduce/template/aiv/kernel/CMakeLists.txt+27-0
@@ -25,4 +25,31 @@ install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_95.o
25 ${INSTALL_OPTIONAL}25 ${INSTALL_OPTIONAL}
26 COMPONENT hccl26 COMPONENT hccl
27)27)
28+ 
29+if(HCCL_ENABLE_910_96)
30+add_custom_target(hccl_aiv_reduce_op_910_96 ALL
31+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_96.o)
32+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_96.o
33+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_reduce_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec
34+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000
35+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
36+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
37+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
38+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
39+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
40+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
41+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
42+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
43+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
44+ COMMAND ${CCE_DIR}/ld.lld -m aicorelinux -Ttext=0 ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_96_tmp.o -static -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_96.o
45+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
46+)
47+ 
48+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_96.o
49+ DESTINATION ${INSTALL_LIBRARY_DIR}
50+ ${INSTALL_OPTIONAL}
51+ COMPONENT hccl
52+)
53+endif()
54+ 
28endif()55endif()
Msrc/ops/reduce_scatter/template/aiv/kernel/CMakeLists.txt+195-0
@@ -188,4 +188,199 @@ install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_bfloat16_v
188 COMPONENT hccl188 COMPONENT hccl
189)189)
190 190 
191+if(HCCL_ENABLE_910_96)
192+add_custom_target(hccl_aiv_reduce_scatter_op_910_96 ALL
193+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_96.o)
194+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_96.o
195+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_reduce_scatter_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec
196+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000
197+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
198+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
199+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
200+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
201+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
202+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
203+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
204+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
205+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
206+ COMMAND ${CCE_DIR}/ld.lld -m aicorelinux -Ttext=0 ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_96_tmp.o -static -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_96.o
207+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
208+)
209+ 
210+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_96.o
211+ DESTINATION ${INSTALL_LIBRARY_DIR}
212+ ${INSTALL_OPTIONAL}
213+ COMPONENT hccl
214+)
215+ 
216+add_custom_target(hccl_rs_superkernel_mesh_1d_int8_v2_910_96 ALL
217+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int8_v2_910_96.o)
218+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int8_v2_910_96.o
219+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int8_v2_910_96.o
220+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
221+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
222+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
223+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
224+ -DHCCL_DTYPE_INT8 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int8
225+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
226+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
227+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
228+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
229+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
230+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
231+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
232+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
233+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
234+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
235+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
236+)
237+ 
238+add_custom_target(hccl_rs_superkernel_mesh_1d_int16_v2_910_96 ALL
239+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int16_v2_910_96.o)
240+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int16_v2_910_96.o
241+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int16_v2_910_96.o
242+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
243+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
244+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
245+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
246+ -DHCCL_DTYPE_INT16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int16
247+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
248+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
249+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
250+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
251+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
252+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
253+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
254+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
255+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
256+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
257+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
258+)
259+ 
260+add_custom_target(hccl_rs_superkernel_mesh_1d_int32_v2_910_96 ALL
261+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int32_v2_910_96.o)
262+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int32_v2_910_96.o
263+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int32_v2_910_96.o
264+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
265+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
266+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
267+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
268+ -DHCCL_DTYPE_INT32 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int32
269+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
270+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
271+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
272+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
273+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
274+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
275+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
276+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
277+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
278+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
279+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
280+)
281+ 
282+add_custom_target(hccl_rs_superkernel_mesh_1d_float16_v2_910_96 ALL
283+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float16_v2_910_96.o)
284+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float16_v2_910_96.o
285+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float16_v2_910_96.o
286+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
287+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
288+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
289+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
290+ -DHCCL_DTYPE_FP16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_float16
291+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
292+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
293+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
294+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
295+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
296+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
297+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
298+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
299+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
300+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
301+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
302+)
303+ 
304+add_custom_target(hccl_rs_superkernel_mesh_1d_float32_v2_910_96 ALL
305+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float32_v2_910_96.o)
306+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float32_v2_910_96.o
307+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float32_v2_910_96.o
308+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
309+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
310+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
311+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
312+ -DHCCL_DTYPE_FP32 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_float32
313+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
314+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
315+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
316+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
317+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
318+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
319+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
320+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
321+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
322+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
323+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
324+)
325+ 
326+add_custom_target(hccl_rs_superkernel_mesh_1d_bfloat16_v2_910_96 ALL
327+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_bfloat16_v2_910_96.o)
328+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_bfloat16_v2_910_96.o
329+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_bfloat16_v2_910_96.o
330+ -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0
331+ -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec
332+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false
333+ -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__
334+ -DHCCL_DTYPE_BFP16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_bfloat16
335+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
336+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
337+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
338+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
339+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
340+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
341+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
342+ -I${CMAKE_CURRENT_SOURCE_DIR}/aiv_interface/
343+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
344+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
345+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
346+)
347+ 
348+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int8_v2_910_96.o
349+ DESTINATION ${INSTALL_LIBRARY_DIR}
350+ ${INSTALL_OPTIONAL}
351+ COMPONENT hccl
352+)
353+ 
354+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int16_v2_910_96.o
355+ DESTINATION ${INSTALL_LIBRARY_DIR}
356+ ${INSTALL_OPTIONAL}
357+ COMPONENT hccl
358+)
359+ 
360+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int32_v2_910_96.o
361+ DESTINATION ${INSTALL_LIBRARY_DIR}
362+ ${INSTALL_OPTIONAL}
363+ COMPONENT hccl
364+)
365+ 
366+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float16_v2_910_96.o
367+ DESTINATION ${INSTALL_LIBRARY_DIR}
368+ ${INSTALL_OPTIONAL}
369+ COMPONENT hccl
370+)
371+ 
372+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float32_v2_910_96.o
373+ DESTINATION ${INSTALL_LIBRARY_DIR}
374+ ${INSTALL_OPTIONAL}
375+ COMPONENT hccl
376+)
377+ 
378+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_bfloat16_v2_910_96.o
379+ DESTINATION ${INSTALL_LIBRARY_DIR}
380+ ${INSTALL_OPTIONAL}
381+ COMPONENT hccl
382+)
383+ 
384+endif()
385+ 
191endif()386endif()
Msrc/ops/scatter/template/aiv/kernel/CMakeLists.txt+27-0
@@ -25,4 +25,31 @@ install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_95.o
25 ${INSTALL_OPTIONAL}25 ${INSTALL_OPTIONAL}
26 COMPONENT hccl26 COMPONENT hccl
27)27)
28+ 
29+if(HCCL_ENABLE_910_96)
30+add_custom_target(hccl_aiv_scatter_op_910_96 ALL
31+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_96.o)
32+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_96.o
33+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_scatter_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec
34+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000
35+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
36+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
37+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
38+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
39+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
40+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
41+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
42+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
43+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
44+ COMMAND ${CCE_DIR}/ld.lld -m aicorelinux -Ttext=0 ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_96_tmp.o -static -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_96.o
45+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
46+)
47+ 
48+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_96.o
49+ DESTINATION ${INSTALL_LIBRARY_DIR}
50+ ${INSTALL_OPTIONAL}
51+ COMPONENT hccl
52+)
53+endif()
54+ 
28endif()55endif()
Msrc/ops/send/template/aiv/kernel/CMakeLists.txt+27-0
@@ -25,4 +25,31 @@ install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_95.o
25 ${INSTALL_OPTIONAL}25 ${INSTALL_OPTIONAL}
26 COMPONENT hccl26 COMPONENT hccl
27)27)
28+ 
29+if(HCCL_ENABLE_910_96)
30+add_custom_target(hccl_aiv_send_op_910_96 ALL
31+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_96.o)
32+add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_96.o
33+ COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_send_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec
34+ --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000
35+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/
36+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw
37+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/interface/
38+ -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/impl/
39+ -I${ASCEND_CANN_PACKAGE_PATH}/include/hccl/
40+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/
41+ -I${OPS_BASE_DIR}/src/ops/op_common/template/aiv/aiv_interface/
42+ -mllvm -cce-aicore-record-overflow=false -mllvm -cce-aicore-addr-transform -mllvm
43+ --cce-aicore-jump-expand=true -std=c++17 -fstack-protector-all
44+ COMMAND ${CCE_DIR}/ld.lld -m aicorelinux -Ttext=0 ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_96_tmp.o -static -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_96.o
45+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
46+)
47+ 
48+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_96.o
49+ DESTINATION ${INSTALL_LIBRARY_DIR}
50+ ${INSTALL_OPTIONAL}
51+ COMPONENT hccl
52+)
53+endif()
54+ 
28endif()55endif()