已合并
add aiv print #2667
yevolin2023创建于 22 天前
add aiv print #2667
已合并
共 26 个文件变更+129-80
| @@ -34,12 +34,18 @@ else() | |||
| 34 | set(ENABLE_DEVICE OFF) | 34 | set(ENABLE_DEVICE OFF) |
| 35 | endif() | 35 | endif() |
| 36 | 36 | ||
| 37 | +# AIV算子调测阶段可配置为1以打开AscendC打印功能 | ||
| 38 | +if(NOT DEFINED ASCENDC_DUMP) | ||
| 39 | + set(ASCENDC_DUMP "0") | ||
| 40 | +endif() | ||
| 41 | + | ||
| 37 | message(STATUS "ASCEND_CANN_PACKAGE_PATH=${ASCEND_CANN_PACKAGE_PATH}") | 42 | message(STATUS "ASCEND_CANN_PACKAGE_PATH=${ASCEND_CANN_PACKAGE_PATH}") |
| 38 | message(STATUS "CANN_3RD_LIB_PATH=${CANN_3RD_LIB_PATH}") | 43 | message(STATUS "CANN_3RD_LIB_PATH=${CANN_3RD_LIB_PATH}") |
| 39 | message(STATUS "PRODUCT_SIDE=${PRODUCT_SIDE}") | 44 | message(STATUS "PRODUCT_SIDE=${PRODUCT_SIDE}") |
| 40 | message(STATUS "BUILD_OPEN_PROJECT=${BUILD_OPEN_PROJECT}") | 45 | message(STATUS "BUILD_OPEN_PROJECT=${BUILD_OPEN_PROJECT}") |
| 41 | message(STATUS "ENABLE_BUILD_DEVICE=${ENABLE_BUILD_DEVICE}") | 46 | message(STATUS "ENABLE_BUILD_DEVICE=${ENABLE_BUILD_DEVICE}") |
| 42 | message(STATUS "ENABLE_BUILD_AARCH=${ENABLE_BUILD_AARCH}") | 47 | message(STATUS "ENABLE_BUILD_AARCH=${ENABLE_BUILD_AARCH}") |
| 48 | +message(STATUS "ASCENDC_DUMP=${ASCENDC_DUMP}") | ||
| 43 | 49 | ||
| 44 | set(HCCL_DIR "${CMAKE_CURRENT_SOURCE_DIR}") | 50 | set(HCCL_DIR "${CMAKE_CURRENT_SOURCE_DIR}") |
| 45 | set(HCC_TOOLCHAIN_DIR "${ASCEND_CANN_PACKAGE_PATH}/toolkit/toolchain/hcc") | 51 | set(HCC_TOOLCHAIN_DIR "${ASCEND_CANN_PACKAGE_PATH}/toolkit/toolchain/hcc") |
| @@ -17,6 +17,7 @@ add_custom_target(hccl_aiv_all_gather_op_910_95 ALL | |||
| 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_95.o) | 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_95.o) |
| 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_95.o | 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_95.o |
| 19 | 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_95_tmp.o --cce-aicore-arch=dav-c310-vec | 19 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_all_gather_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_95_tmp.o --cce-aicore-arch=dav-c310-vec |
| 20 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 20 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 21 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 21 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 22 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| 22 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 23 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| @@ -41,7 +42,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_int8_v2 ALL | |||
| 41 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int8_v2.o | 42 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int8_v2.o |
| 42 | 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.o | 43 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int8_v2.o |
| 43 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 44 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 44 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 45 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 45 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 46 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 46 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 47 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 47 | -DHCCL_DTYPE_INT8 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int8 | 48 | -DHCCL_DTYPE_INT8 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int8 |
| @@ -62,7 +63,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_int16_v2 ALL | |||
| 62 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int16_v2.o | 63 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int16_v2.o |
| 63 | 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.o | 64 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int16_v2.o |
| 64 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 65 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 65 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 66 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 66 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 67 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 67 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 68 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 68 | -DHCCL_DTYPE_INT16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int16 | 69 | -DHCCL_DTYPE_INT16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int16 |
| @@ -83,7 +84,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_int32_v2 ALL | |||
| 83 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int32_v2.o | 84 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int32_v2.o |
| 84 | 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.o | 85 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int32_v2.o |
| 85 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 86 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 86 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 87 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 87 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 88 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 88 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 89 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 89 | -DHCCL_DTYPE_INT32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int32 | 90 | -DHCCL_DTYPE_INT32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int32 |
| @@ -104,7 +105,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_float16_v2 ALL | |||
| 104 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float16_v2.o | 105 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float16_v2.o |
| 105 | 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.o | 106 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float16_v2.o |
| 106 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 107 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 107 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 108 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 108 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 109 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 109 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 110 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 110 | -DHCCL_DTYPE_FP16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_float16 | 111 | -DHCCL_DTYPE_FP16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_float16 |
| @@ -125,7 +126,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_float32_v2 ALL | |||
| 125 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float32_v2.o | 126 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float32_v2.o |
| 126 | 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.o | 127 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float32_v2.o |
| 127 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 128 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 128 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 129 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 129 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 130 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 130 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 131 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 131 | -DHCCL_DTYPE_FP32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_float32 | 132 | -DHCCL_DTYPE_FP32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_float32 |
| @@ -146,7 +147,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_bfloat16_v2 ALL | |||
| 146 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_bfloat16_v2.o | 147 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_bfloat16_v2.o |
| 147 | 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.o | 148 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_bfloat16_v2.o |
| 148 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 149 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 149 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 150 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 150 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 151 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 151 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 152 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 152 | -DHCCL_DTYPE_BFP16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_bfloat16 | 153 | -DHCCL_DTYPE_BFP16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_bfloat16 |
| @@ -167,7 +168,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_uint8_v2 ALL | |||
| 167 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint8_v2.o | 168 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint8_v2.o |
| 168 | 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.o | 169 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint8_v2.o |
| 169 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 170 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 170 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 171 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 171 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 172 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 172 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 173 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 173 | -DHCCL_DTYPE_UINT8 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint8 | 174 | -DHCCL_DTYPE_UINT8 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint8 |
| @@ -188,7 +189,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_uint16_v2 ALL | |||
| 188 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint16_v2.o | 189 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint16_v2.o |
| 189 | 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.o | 190 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint16_v2.o |
| 190 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 191 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 191 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 192 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 192 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 193 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 193 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 194 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 194 | -DHCCL_DTYPE_UINT16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint16 | 195 | -DHCCL_DTYPE_UINT16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint16 |
| @@ -209,7 +210,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_uint32_v2 ALL | |||
| 209 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint32_v2.o | 210 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint32_v2.o |
| 210 | 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.o | 211 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint32_v2.o |
| 211 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 212 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 212 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 213 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 213 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 214 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 214 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 215 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 215 | -DHCCL_DTYPE_UINT32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint32 | 216 | -DHCCL_DTYPE_UINT32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint32 |
| @@ -230,7 +231,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_int64_v2 ALL | |||
| 230 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int64_v2.o | 231 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int64_v2.o |
| 231 | 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.o | 232 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int64_v2.o |
| 232 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 233 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 233 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 234 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 234 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 235 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 235 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 236 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 236 | -DHCCL_DTYPE_INT64 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int64 | 237 | -DHCCL_DTYPE_INT64 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int64 |
| @@ -251,7 +252,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_uint64_v2 ALL | |||
| 251 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint64_v2.o | 252 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint64_v2.o |
| 252 | 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.o | 253 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint64_v2.o |
| 253 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 254 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 254 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 255 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 255 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 256 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 256 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 257 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 257 | -DHCCL_DTYPE_UINT64 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint64 | 258 | -DHCCL_DTYPE_UINT64 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint64 |
| @@ -338,6 +339,7 @@ add_custom_target(hccl_aiv_all_gather_op_910_96 ALL | |||
| 338 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_96.o) | 339 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_96.o) |
| 339 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_96.o | 340 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_96.o |
| 340 | 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 | 341 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_all_gather_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_gather_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec |
| 342 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 341 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 343 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 342 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ | 344 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ |
| 343 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw | 345 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw |
| @@ -363,7 +365,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_int8_v2_910_96 ALL | |||
| 363 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int8_v2_910_96.o | 365 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int8_v2_910_96.o |
| 364 | 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 | 366 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int8_v2_910_96.o |
| 365 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 367 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 366 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 368 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 367 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 369 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 368 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 370 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 369 | -DHCCL_DTYPE_INT8 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int8 | 371 | -DHCCL_DTYPE_INT8 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int8 |
| @@ -391,7 +393,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_int16_v2_910_96 ALL | |||
| 391 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int16_v2_910_96.o | 393 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int16_v2_910_96.o |
| 392 | 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 | 394 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int16_v2_910_96.o |
| 393 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 395 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 394 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 396 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 395 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 397 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 396 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 398 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 397 | -DHCCL_DTYPE_INT16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int16 | 399 | -DHCCL_DTYPE_INT16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int16 |
| @@ -419,7 +421,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_int32_v2_910_96 ALL | |||
| 419 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int32_v2_910_96.o | 421 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int32_v2_910_96.o |
| 420 | 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 | 422 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int32_v2_910_96.o |
| 421 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 423 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 422 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 424 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 423 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 425 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 424 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 426 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 425 | -DHCCL_DTYPE_INT32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int32 | 427 | -DHCCL_DTYPE_INT32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int32 |
| @@ -447,7 +449,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_float16_v2_910_96 ALL | |||
| 447 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float16_v2_910_96.o | 449 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float16_v2_910_96.o |
| 448 | 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 | 450 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float16_v2_910_96.o |
| 449 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 451 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 450 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 452 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 451 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 453 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 452 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 454 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 453 | -DHCCL_DTYPE_FP16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_float16 | 455 | -DHCCL_DTYPE_FP16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_float16 |
| @@ -475,7 +477,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_float32_v2_910_96 ALL | |||
| 475 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float32_v2_910_96.o | 477 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float32_v2_910_96.o |
| 476 | 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 | 478 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_float32_v2_910_96.o |
| 477 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 479 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 478 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 480 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 479 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 481 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 480 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 482 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 481 | -DHCCL_DTYPE_FP32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_float32 | 483 | -DHCCL_DTYPE_FP32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_float32 |
| @@ -503,7 +505,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_bfloat16_v2_910_96 ALL | |||
| 503 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_bfloat16_v2_910_96.o | 505 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_bfloat16_v2_910_96.o |
| 504 | 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 | 506 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_bfloat16_v2_910_96.o |
| 505 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 507 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 506 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 508 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 507 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 509 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 508 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 510 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 509 | -DHCCL_DTYPE_BFP16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_bfloat16 | 511 | -DHCCL_DTYPE_BFP16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_bfloat16 |
| @@ -531,7 +533,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_uint8_v2_910_96 ALL | |||
| 531 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint8_v2_910_96.o | 533 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint8_v2_910_96.o |
| 532 | 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 | 534 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint8_v2_910_96.o |
| 533 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 535 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 534 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 536 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 535 | --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 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 536 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 538 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 537 | -DHCCL_DTYPE_UINT8 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint8 | 539 | -DHCCL_DTYPE_UINT8 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint8 |
| @@ -559,7 +561,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_uint16_v2_910_96 ALL | |||
| 559 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint16_v2_910_96.o | 561 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint16_v2_910_96.o |
| 560 | 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 | 562 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint16_v2_910_96.o |
| 561 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 563 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 562 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 564 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 563 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 565 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 564 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 566 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 565 | -DHCCL_DTYPE_UINT16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint16 | 567 | -DHCCL_DTYPE_UINT16 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint16 |
| @@ -587,7 +589,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_uint32_v2_910_96 ALL | |||
| 587 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint32_v2_910_96.o | 589 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint32_v2_910_96.o |
| 588 | 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 | 590 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint32_v2_910_96.o |
| 589 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 591 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 590 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 592 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 591 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 593 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 592 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 594 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 593 | -DHCCL_DTYPE_UINT32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint32 | 595 | -DHCCL_DTYPE_UINT32 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint32 |
| @@ -615,7 +617,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_int64_v2_910_96 ALL | |||
| 615 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int64_v2_910_96.o | 617 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int64_v2_910_96.o |
| 616 | 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 | 618 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_int64_v2_910_96.o |
| 617 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 619 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 618 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 620 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 619 | --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 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 620 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 622 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 621 | -DHCCL_DTYPE_INT64 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int64 | 623 | -DHCCL_DTYPE_INT64 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_int64 |
| @@ -643,7 +645,7 @@ add_custom_target(hccl_ag_superkernel_mesh_1d_uint64_v2_910_96 ALL | |||
| 643 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint64_v2_910_96.o | 645 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint64_v2_910_96.o |
| 644 | 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 | 646 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ag_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ag_superkernel_mesh_1d_uint64_v2_910_96.o |
| 645 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 647 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 646 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 648 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 647 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 649 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 648 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 650 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 649 | -DHCCL_DTYPE_UINT64 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint64 | 651 | -DHCCL_DTYPE_UINT64 -Dsk_allgather_mesh_1d=sk_allgather_mesh_1d_uint64 |
| @@ -13,6 +13,7 @@ | |||
| 13 | 13 | ||
| 14 | extern "C" __aicore__ void sk_allgather_mesh_1d(SUPERKERNEL_LITE_ARGS_DEF) | 14 | extern "C" __aicore__ void sk_allgather_mesh_1d(SUPERKERNEL_LITE_ARGS_DEF) |
| 15 | { | 15 | { |
| 16 | + AIV_INFO_HINT; | ||
| 16 | SUPERKERNEL_LITE_ARGS_EXTRACT; | 17 | SUPERKERNEL_LITE_ARGS_EXTRACT; |
| 17 | return sk_ag_mesh_1d(SUPERKERNEL_ARGS_CALL); | 18 | return sk_ag_mesh_1d(SUPERKERNEL_ARGS_CALL); |
| 18 | } | 19 | } |
| @@ -19,6 +19,7 @@ using namespace AscendC; | |||
| 19 | 19 | ||
| 20 | extern "C" __global__ __aicore__ void aiv_all_gather_# | 20 | extern "C" __global__ __aicore__ void aiv_all_gather_# |
| 21 | { \ | 21 | { \ |
| 22 | + AIV_INFO_HINT; \ | ||
| 22 | return AivAllGatherV2Mesh1D<type>(KERNEL_ARGS_CALL); \ | 23 | return AivAllGatherV2Mesh1D<type>(KERNEL_ARGS_CALL); \ |
| 23 | } \ | 24 | } \ |
| 24 | EXPORT_AIV_META_INFO(aiv_all_gather_# | 25 | EXPORT_AIV_META_INFO(aiv_all_gather_# |
| @@ -17,6 +17,7 @@ add_custom_target(hccl_aiv_all_reduce_op_910_95 ALL | |||
| 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_95.o) | 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_95.o) |
| 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_95.o | 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_95.o |
| 19 | 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_95_tmp.o --cce-aicore-arch=dav-c310-vec | 19 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_all_reduce_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_95_tmp.o --cce-aicore-arch=dav-c310-vec |
| 20 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 20 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 21 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 21 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 22 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| 22 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 23 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| @@ -41,7 +42,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_int8_v2 ALL | |||
| 41 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int8_v2.o | 42 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int8_v2.o |
| 42 | 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.o | 43 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int8_v2.o |
| 43 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 44 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 44 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 45 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 45 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 46 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 46 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 47 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 47 | -DHCCL_DTYPE_INT8 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int8 | 48 | -DHCCL_DTYPE_INT8 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int8 |
| @@ -62,7 +63,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_int16_v2 ALL | |||
| 62 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int16_v2.o | 63 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int16_v2.o |
| 63 | 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.o | 64 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int16_v2.o |
| 64 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 65 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 65 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 66 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 66 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 67 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 67 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 68 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 68 | -DHCCL_DTYPE_INT16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int16 | 69 | -DHCCL_DTYPE_INT16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int16 |
| @@ -83,7 +84,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_int32_v2 ALL | |||
| 83 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int32_v2.o | 84 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int32_v2.o |
| 84 | 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.o | 85 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int32_v2.o |
| 85 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 86 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 86 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 87 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 87 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 88 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 88 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 89 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 89 | -DHCCL_DTYPE_INT32 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int32 | 90 | -DHCCL_DTYPE_INT32 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int32 |
| @@ -104,7 +105,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_float16_v2 ALL | |||
| 104 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float16_v2.o | 105 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float16_v2.o |
| 105 | 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.o | 106 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float16_v2.o |
| 106 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 107 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 107 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 108 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 108 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 109 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 109 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 110 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 110 | -DHCCL_DTYPE_FP16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_float16 | 111 | -DHCCL_DTYPE_FP16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_float16 |
| @@ -125,7 +126,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_float32_v2 ALL | |||
| 125 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float32_v2.o | 126 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float32_v2.o |
| 126 | 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.o | 127 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float32_v2.o |
| 127 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 128 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 128 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 129 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 129 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 130 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 130 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 131 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 131 | -DHCCL_DTYPE_FP32 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_float32 | 132 | -DHCCL_DTYPE_FP32 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_float32 |
| @@ -146,7 +147,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2 ALL | |||
| 146 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2.o | 147 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2.o |
| 147 | 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.o | 148 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2.o |
| 148 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 149 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 149 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 150 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 150 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 151 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 151 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 152 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 152 | -DHCCL_DTYPE_BFP16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_bfloat16 | 153 | -DHCCL_DTYPE_BFP16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_bfloat16 |
| @@ -167,7 +168,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_int8_v2 ALL | |||
| 167 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int8_v2.o | 168 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int8_v2.o |
| 168 | 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.o | 169 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int8_v2.o |
| 169 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 170 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 170 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 171 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 171 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 172 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 172 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 173 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 173 | -DHCCL_DTYPE_INT8 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int8 | 174 | -DHCCL_DTYPE_INT8 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int8 |
| @@ -188,7 +189,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_int16_v2 ALL | |||
| 188 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int16_v2.o | 189 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int16_v2.o |
| 189 | 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.o | 190 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int16_v2.o |
| 190 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 191 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 191 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 192 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 192 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 193 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 193 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 194 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 194 | -DHCCL_DTYPE_INT16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int16 | 195 | -DHCCL_DTYPE_INT16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int16 |
| @@ -209,7 +210,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_int32_v2 ALL | |||
| 209 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int32_v2.o | 210 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int32_v2.o |
| 210 | 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.o | 211 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int32_v2.o |
| 211 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 212 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 212 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 213 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 213 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 214 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 214 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 215 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 215 | -DHCCL_DTYPE_INT32 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int32 | 216 | -DHCCL_DTYPE_INT32 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int32 |
| @@ -230,7 +231,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_float16_v2 ALL | |||
| 230 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float16_v2.o | 231 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float16_v2.o |
| 231 | 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.o | 232 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float16_v2.o |
| 232 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 233 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 233 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 234 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 234 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 235 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 235 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 236 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 236 | -DHCCL_DTYPE_FP16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_float16 | 237 | -DHCCL_DTYPE_FP16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_float16 |
| @@ -251,7 +252,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_float32_v2 ALL | |||
| 251 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float32_v2.o | 252 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float32_v2.o |
| 252 | 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.o | 253 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float32_v2.o |
| 253 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 254 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 254 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 255 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 255 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 256 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 256 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 257 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 257 | -DHCCL_DTYPE_FP32 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_float32 | 258 | -DHCCL_DTYPE_FP32 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_float32 |
| @@ -272,7 +273,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2 ALL | |||
| 272 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2.o | 273 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2.o |
| 273 | 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.o | 274 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2.o |
| 274 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 275 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 275 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 276 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 276 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 277 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 277 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 278 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 278 | -DHCCL_DTYPE_BFP16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_bfloat16 | 279 | -DHCCL_DTYPE_BFP16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_bfloat16 |
| @@ -365,6 +366,7 @@ add_custom_target(hccl_aiv_all_reduce_op_910_96 ALL | |||
| 365 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_96.o) | 366 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_96.o) |
| 366 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_96.o | 367 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_96.o |
| 367 | 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 | 368 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_all_reduce_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_reduce_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec |
| 369 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 368 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 370 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 369 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ | 371 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ |
| 370 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw | 372 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw |
| @@ -384,7 +386,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_int8_v2_910_96 ALL | |||
| 384 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int8_v2_910_96.o | 386 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int8_v2_910_96.o |
| 385 | 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 | 387 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int8_v2_910_96.o |
| 386 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 388 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 387 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 389 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 388 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 390 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 389 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 391 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 390 | -DHCCL_DTYPE_INT8 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int8 | 392 | -DHCCL_DTYPE_INT8 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int8 |
| @@ -406,7 +408,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_int16_v2_910_96 ALL | |||
| 406 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int16_v2_910_96.o | 408 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int16_v2_910_96.o |
| 407 | 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 | 409 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int16_v2_910_96.o |
| 408 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 410 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 409 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 411 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 410 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 412 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 411 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 413 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 412 | -DHCCL_DTYPE_INT16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int16 | 414 | -DHCCL_DTYPE_INT16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int16 |
| @@ -428,7 +430,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_int32_v2_910_96 ALL | |||
| 428 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int32_v2_910_96.o | 430 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int32_v2_910_96.o |
| 429 | 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 | 431 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_int32_v2_910_96.o |
| 430 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 432 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 431 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 433 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 432 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 434 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 433 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 435 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 434 | -DHCCL_DTYPE_INT32 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int32 | 436 | -DHCCL_DTYPE_INT32 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_int32 |
| @@ -450,7 +452,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_float16_v2_910_96 ALL | |||
| 450 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float16_v2_910_96.o | 452 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float16_v2_910_96.o |
| 451 | 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 | 453 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float16_v2_910_96.o |
| 452 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 454 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 453 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 455 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 454 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 456 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 455 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 457 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 456 | -DHCCL_DTYPE_FP16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_float16 | 458 | -DHCCL_DTYPE_FP16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_float16 |
| @@ -472,7 +474,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_float32_v2_910_96 ALL | |||
| 472 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float32_v2_910_96.o | 474 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float32_v2_910_96.o |
| 473 | 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 | 475 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_float32_v2_910_96.o |
| 474 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 476 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 475 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 477 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 476 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 478 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 477 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 479 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 478 | -DHCCL_DTYPE_FP32 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_float32 | 480 | -DHCCL_DTYPE_FP32 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_float32 |
| @@ -494,7 +496,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2_910_96 ALL | |||
| 494 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2_910_96.o | 496 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2_910_96.o |
| 495 | 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 | 497 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_oneshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_oneshot_bfloat16_v2_910_96.o |
| 496 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 498 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 497 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 499 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 498 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 500 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 499 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 501 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 500 | -DHCCL_DTYPE_BFP16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_bfloat16 | 502 | -DHCCL_DTYPE_BFP16 -Dsk_allreduce_mesh_1d_oneshot=sk_allreduce_mesh_1d_oneshot_bfloat16 |
| @@ -516,7 +518,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_int8_v2_910_96 ALL | |||
| 516 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int8_v2_910_96.o | 518 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int8_v2_910_96.o |
| 517 | 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 | 519 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int8_v2_910_96.o |
| 518 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 520 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 519 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 521 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 520 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 522 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 521 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 523 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 522 | -DHCCL_DTYPE_INT8 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int8 | 524 | -DHCCL_DTYPE_INT8 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int8 |
| @@ -538,7 +540,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_int16_v2_910_96 ALL | |||
| 538 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int16_v2_910_96.o | 540 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int16_v2_910_96.o |
| 539 | 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 | 541 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int16_v2_910_96.o |
| 540 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 542 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 541 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 543 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 542 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 544 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 543 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 545 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 544 | -DHCCL_DTYPE_INT16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int16 | 546 | -DHCCL_DTYPE_INT16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int16 |
| @@ -560,7 +562,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_int32_v2_910_96 ALL | |||
| 560 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int32_v2_910_96.o | 562 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int32_v2_910_96.o |
| 561 | 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 | 563 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_int32_v2_910_96.o |
| 562 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 564 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 563 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 565 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 564 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 566 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 565 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 567 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 566 | -DHCCL_DTYPE_INT32 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int32 | 568 | -DHCCL_DTYPE_INT32 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_int32 |
| @@ -582,7 +584,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_float16_v2_910_96 ALL | |||
| 582 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float16_v2_910_96.o | 584 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float16_v2_910_96.o |
| 583 | 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 | 585 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float16_v2_910_96.o |
| 584 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 586 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 585 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 587 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 586 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 588 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 587 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 589 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 588 | -DHCCL_DTYPE_FP16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_float16 | 590 | -DHCCL_DTYPE_FP16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_float16 |
| @@ -604,7 +606,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_float32_v2_910_96 ALL | |||
| 604 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float32_v2_910_96.o | 606 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float32_v2_910_96.o |
| 605 | 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 | 607 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_float32_v2_910_96.o |
| 606 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 608 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 607 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 609 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 608 | --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 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 609 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 611 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 610 | -DHCCL_DTYPE_FP32 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_float32 | 612 | -DHCCL_DTYPE_FP32 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_float32 |
| @@ -626,7 +628,7 @@ add_custom_target(hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2_910_96 ALL | |||
| 626 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2_910_96.o | 628 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2_910_96.o |
| 627 | 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 | 629 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_ar_mesh_1d_twoshot_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_ar_superkernel_mesh_1d_twoshot_bfloat16_v2_910_96.o |
| 628 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 630 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 629 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 631 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 630 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 632 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 631 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 633 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 632 | -DHCCL_DTYPE_BFP16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_bfloat16 | 634 | -DHCCL_DTYPE_BFP16 -Dsk_allreduce_mesh_1d_twoshot=sk_allreduce_mesh_1d_twoshot_bfloat16 |
| @@ -20,6 +20,7 @@ using namespace AscendC; | |||
| 20 | 20 | ||
| 21 | extern "C" __global__ __aicore__ void aiv_allreduce_# | 21 | extern "C" __global__ __aicore__ void aiv_allreduce_# |
| 22 | { \ | 22 | { \ |
| 23 | + AIV_INFO_HINT; \ | ||
| 23 | return AivAllReduceV2Mesh1DOneShot<type>(KERNEL_ARGS_CALL); \ | 24 | return AivAllReduceV2Mesh1DOneShot<type>(KERNEL_ARGS_CALL); \ |
| 24 | } \ | 25 | } \ |
| 25 | EXPORT_AIV_META_INFO(aiv_allreduce_# | 26 | EXPORT_AIV_META_INFO(aiv_allreduce_# |
| @@ -27,6 +28,7 @@ using namespace AscendC; | |||
| 27 | 28 | ||
| 28 | extern "C" __global__ __aicore__ void aiv_allreduce_mesh1d_twoshot_# | 29 | extern "C" __global__ __aicore__ void aiv_allreduce_mesh1d_twoshot_# |
| 29 | { \ | 30 | { \ |
| 31 | + AIV_INFO_HINT; \ | ||
| 30 | return AivAllReduceV2Mesh1DTwoShot<type>(KERNEL_ARGS_CALL); \ | 32 | return AivAllReduceV2Mesh1DTwoShot<type>(KERNEL_ARGS_CALL); \ |
| 31 | } \ | 33 | } \ |
| 32 | EXPORT_AIV_META_INFO(aiv_allreduce_mesh1d_twoshot_# | 34 | EXPORT_AIV_META_INFO(aiv_allreduce_mesh1d_twoshot_# |
| @@ -13,6 +13,7 @@ | |||
| 13 | 13 | ||
| 14 | extern "C" __aicore__ void sk_allreduce_mesh_1d_oneshot(SUPERKERNEL_LITE_ARGS_DEF) | 14 | extern "C" __aicore__ void sk_allreduce_mesh_1d_oneshot(SUPERKERNEL_LITE_ARGS_DEF) |
| 15 | { | 15 | { |
| 16 | + AIV_INFO_HINT; | ||
| 16 | SUPERKERNEL_LITE_ARGS_EXTRACT; | 17 | SUPERKERNEL_LITE_ARGS_EXTRACT; |
| 17 | return sk_ar_mesh_1d_oneshot(SUPERKERNEL_ARGS_CALL); | 18 | return sk_ar_mesh_1d_oneshot(SUPERKERNEL_ARGS_CALL); |
| 18 | } | 19 | } |
| @@ -13,6 +13,7 @@ | |||
| 13 | 13 | ||
| 14 | extern "C" __aicore__ void sk_allreduce_mesh_1d_twoshot(SUPERKERNEL_LITE_ARGS_DEF) | 14 | extern "C" __aicore__ void sk_allreduce_mesh_1d_twoshot(SUPERKERNEL_LITE_ARGS_DEF) |
| 15 | { | 15 | { |
| 16 | + AIV_INFO_HINT; | ||
| 16 | SUPERKERNEL_LITE_ARGS_EXTRACT; | 17 | SUPERKERNEL_LITE_ARGS_EXTRACT; |
| 17 | return sk_ar_mesh_1d_twoshot(SUPERKERNEL_ARGS_CALL); | 18 | return sk_ar_mesh_1d_twoshot(SUPERKERNEL_ARGS_CALL); |
| 18 | } | 19 | } |
| @@ -17,6 +17,7 @@ add_custom_target(hccl_aiv_all_to_all_v_op_910_95 ALL | |||
| 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_95.o) | 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_95.o) |
| 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_95.o | 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_95.o |
| 19 | 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_95_tmp.o --cce-aicore-arch=dav-c310-vec | 19 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_all_to_all_v_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_95_tmp.o --cce-aicore-arch=dav-c310-vec |
| 20 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 20 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 21 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 21 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 22 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| 22 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 23 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| @@ -34,6 +35,7 @@ add_custom_target(hccl_aiv_all_to_all_op_910_95 ALL | |||
| 34 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_95.o) | 35 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_95.o) |
| 35 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_95.o | 36 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_95.o |
| 36 | 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_95_tmp.o --cce-aicore-arch=dav-c310-vec | 37 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_all_to_all_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_95_tmp.o --cce-aicore-arch=dav-c310-vec |
| 38 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 37 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 39 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 38 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 40 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| 39 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 41 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| @@ -64,7 +66,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_int8_v2 ALL | |||
| 64 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int8_v2.o | 66 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int8_v2.o |
| 65 | 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.o | 67 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int8_v2.o |
| 66 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 68 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 67 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 69 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 68 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 70 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 69 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 71 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 70 | -DHCCL_DTYPE_INT8 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int8 | 72 | -DHCCL_DTYPE_INT8 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int8 |
| @@ -85,7 +87,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_int16_v2 ALL | |||
| 85 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int16_v2.o | 87 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int16_v2.o |
| 86 | 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.o | 88 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int16_v2.o |
| 87 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 89 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 88 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 90 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 89 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 91 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 90 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 92 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 91 | -DHCCL_DTYPE_INT16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int16 | 93 | -DHCCL_DTYPE_INT16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int16 |
| @@ -106,7 +108,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_int32_v2 ALL | |||
| 106 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int32_v2.o | 108 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int32_v2.o |
| 107 | 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.o | 109 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int32_v2.o |
| 108 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 110 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 109 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 111 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 110 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 112 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 111 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 113 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 112 | -DHCCL_DTYPE_INT32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int32 | 114 | -DHCCL_DTYPE_INT32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int32 |
| @@ -127,7 +129,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_float16_v2 ALL | |||
| 127 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float16_v2.o | 129 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float16_v2.o |
| 128 | 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.o | 130 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float16_v2.o |
| 129 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 131 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 130 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 132 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 131 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 133 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 132 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 134 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 133 | -DHCCL_DTYPE_FP16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_float16 | 135 | -DHCCL_DTYPE_FP16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_float16 |
| @@ -148,7 +150,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_float32_v2 ALL | |||
| 148 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float32_v2.o | 150 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float32_v2.o |
| 149 | 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.o | 151 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float32_v2.o |
| 150 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 152 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 151 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 153 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 152 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 154 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 153 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 155 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 154 | -DHCCL_DTYPE_FP32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_float32 | 156 | -DHCCL_DTYPE_FP32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_float32 |
| @@ -169,7 +171,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_bfloat16_v2 ALL | |||
| 169 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_bfloat16_v2.o | 171 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_bfloat16_v2.o |
| 170 | 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.o | 172 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_bfloat16_v2.o |
| 171 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 173 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 172 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 174 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 173 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 175 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 174 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 176 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 175 | -DHCCL_DTYPE_BFP16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_bfloat16 | 177 | -DHCCL_DTYPE_BFP16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_bfloat16 |
| @@ -190,7 +192,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_uint8_v2 ALL | |||
| 190 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint8_v2.o | 192 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint8_v2.o |
| 191 | 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.o | 193 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint8_v2.o |
| 192 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 194 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 193 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 195 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 194 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 196 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 195 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 197 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 196 | -DHCCL_DTYPE_UINT8 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint8 | 198 | -DHCCL_DTYPE_UINT8 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint8 |
| @@ -211,7 +213,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_uint16_v2 ALL | |||
| 211 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint16_v2.o | 213 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint16_v2.o |
| 212 | 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.o | 214 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint16_v2.o |
| 213 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 215 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 214 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 216 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 215 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 217 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 216 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 218 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 217 | -DHCCL_DTYPE_UINT16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint16 | 219 | -DHCCL_DTYPE_UINT16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint16 |
| @@ -232,7 +234,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_uint32_v2 ALL | |||
| 232 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint32_v2.o | 234 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint32_v2.o |
| 233 | 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.o | 235 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint32_v2.o |
| 234 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 236 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 235 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 237 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 236 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 238 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 237 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 239 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 238 | -DHCCL_DTYPE_UINT32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint32 | 240 | -DHCCL_DTYPE_UINT32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint32 |
| @@ -253,7 +255,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_int64_v2 ALL | |||
| 253 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int64_v2.o | 255 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int64_v2.o |
| 254 | 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.o | 256 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int64_v2.o |
| 255 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 257 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 256 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 258 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 257 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 259 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 258 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 260 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 259 | -DHCCL_DTYPE_INT64 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int64 | 261 | -DHCCL_DTYPE_INT64 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int64 |
| @@ -274,7 +276,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_uint64_v2 ALL | |||
| 274 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint64_v2.o | 276 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint64_v2.o |
| 275 | 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.o | 277 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint64_v2.o |
| 276 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 278 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 277 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 279 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 278 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 280 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 279 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 281 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 280 | -DHCCL_DTYPE_UINT64 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint64 | 282 | -DHCCL_DTYPE_UINT64 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint64 |
| @@ -361,6 +363,7 @@ add_custom_target(hccl_aiv_all_to_all_v_op_910_96 ALL | |||
| 361 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_96.o) | 363 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_96.o) |
| 362 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_96.o | 364 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_96.o |
| 363 | 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 | 365 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_all_to_all_v_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_v_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec |
| 366 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 364 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 367 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 365 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ | 368 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ |
| 366 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw | 369 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw |
| @@ -379,6 +382,7 @@ add_custom_target(hccl_aiv_all_to_all_op_910_96 ALL | |||
| 379 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_96.o) | 382 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_96.o) |
| 380 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_96.o | 383 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_96.o |
| 381 | 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 | 384 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_all_to_all_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_all_to_all_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec |
| 385 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 382 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 386 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 383 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ | 387 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ |
| 384 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw | 388 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw |
| @@ -410,7 +414,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_int8_v2_910_96 ALL | |||
| 410 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int8_v2_910_96.o | 414 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int8_v2_910_96.o |
| 411 | 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 | 415 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int8_v2_910_96.o |
| 412 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 416 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 413 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 417 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 414 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 418 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 415 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 419 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 416 | -DHCCL_DTYPE_INT8 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int8 | 420 | -DHCCL_DTYPE_INT8 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int8 |
| @@ -432,7 +436,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_int16_v2_910_96 ALL | |||
| 432 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int16_v2_910_96.o | 436 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int16_v2_910_96.o |
| 433 | 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 | 437 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int16_v2_910_96.o |
| 434 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 438 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 435 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 439 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 436 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 440 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 437 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 441 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 438 | -DHCCL_DTYPE_INT16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int16 | 442 | -DHCCL_DTYPE_INT16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int16 |
| @@ -454,7 +458,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_int32_v2_910_96 ALL | |||
| 454 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int32_v2_910_96.o | 458 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int32_v2_910_96.o |
| 455 | 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 | 459 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int32_v2_910_96.o |
| 456 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 460 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 457 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 461 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 458 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 462 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 459 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 463 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 460 | -DHCCL_DTYPE_INT32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int32 | 464 | -DHCCL_DTYPE_INT32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int32 |
| @@ -476,7 +480,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_float16_v2_910_96 ALL | |||
| 476 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float16_v2_910_96.o | 480 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float16_v2_910_96.o |
| 477 | 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 | 481 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float16_v2_910_96.o |
| 478 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 482 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 479 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 483 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 480 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 484 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 481 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 485 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 482 | -DHCCL_DTYPE_FP16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_float16 | 486 | -DHCCL_DTYPE_FP16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_float16 |
| @@ -498,7 +502,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_float32_v2_910_96 ALL | |||
| 498 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float32_v2_910_96.o | 502 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float32_v2_910_96.o |
| 499 | 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 | 503 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_float32_v2_910_96.o |
| 500 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 504 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 501 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 505 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 502 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 506 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 503 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 507 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 504 | -DHCCL_DTYPE_FP32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_float32 | 508 | -DHCCL_DTYPE_FP32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_float32 |
| @@ -520,7 +524,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_bfloat16_v2_910_96 ALL | |||
| 520 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_bfloat16_v2_910_96.o | 524 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_bfloat16_v2_910_96.o |
| 521 | 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 | 525 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_bfloat16_v2_910_96.o |
| 522 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 526 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 523 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 527 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 524 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 528 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 525 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 529 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 526 | -DHCCL_DTYPE_BFP16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_bfloat16 | 530 | -DHCCL_DTYPE_BFP16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_bfloat16 |
| @@ -542,7 +546,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_uint8_v2_910_96 ALL | |||
| 542 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint8_v2_910_96.o | 546 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint8_v2_910_96.o |
| 543 | 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 | 547 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint8_v2_910_96.o |
| 544 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 548 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 545 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 549 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 546 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 550 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 547 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 551 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 548 | -DHCCL_DTYPE_UINT8 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint8 | 552 | -DHCCL_DTYPE_UINT8 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint8 |
| @@ -564,7 +568,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_uint16_v2_910_96 ALL | |||
| 564 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint16_v2_910_96.o | 568 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint16_v2_910_96.o |
| 565 | 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 | 569 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint16_v2_910_96.o |
| 566 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 570 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 567 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 571 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 568 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 572 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 569 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 573 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 570 | -DHCCL_DTYPE_UINT16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint16 | 574 | -DHCCL_DTYPE_UINT16 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint16 |
| @@ -586,7 +590,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_uint32_v2_910_96 ALL | |||
| 586 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint32_v2_910_96.o | 590 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint32_v2_910_96.o |
| 587 | 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 | 591 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint32_v2_910_96.o |
| 588 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 592 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 589 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 593 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 590 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 594 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 591 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 595 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 592 | -DHCCL_DTYPE_UINT32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint32 | 596 | -DHCCL_DTYPE_UINT32 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint32 |
| @@ -608,7 +612,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_int64_v2_910_96 ALL | |||
| 608 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int64_v2_910_96.o | 612 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int64_v2_910_96.o |
| 609 | 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 | 613 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_int64_v2_910_96.o |
| 610 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 614 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 611 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 615 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 612 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 616 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 613 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 617 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 614 | -DHCCL_DTYPE_INT64 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int64 | 618 | -DHCCL_DTYPE_INT64 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_int64 |
| @@ -630,7 +634,7 @@ add_custom_target(hccl_a2a_superkernel_mesh_1d_uint64_v2_910_96 ALL | |||
| 630 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint64_v2_910_96.o | 634 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint64_v2_910_96.o |
| 631 | 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 | 635 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_a2a_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_a2a_superkernel_mesh_1d_uint64_v2_910_96.o |
| 632 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 636 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 633 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 637 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 634 | --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 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 635 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 639 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 636 | -DHCCL_DTYPE_UINT64 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint64 | 640 | -DHCCL_DTYPE_UINT64 -Dsk_alltoall_mesh_1d=sk_alltoall_mesh_1d_uint64 |
| @@ -13,6 +13,7 @@ | |||
| 13 | 13 | ||
| 14 | extern "C" __aicore__ void sk_alltoall_mesh_1d(SUPERKERNEL_LITE_ARGS_DEF) | 14 | extern "C" __aicore__ void sk_alltoall_mesh_1d(SUPERKERNEL_LITE_ARGS_DEF) |
| 15 | { | 15 | { |
| 16 | + AIV_INFO_HINT; | ||
| 16 | SUPERKERNEL_LITE_ARGS_EXTRACT; | 17 | SUPERKERNEL_LITE_ARGS_EXTRACT; |
| 17 | return sk_a2a_mesh_1d(SUPERKERNEL_ARGS_CALL); | 18 | return sk_a2a_mesh_1d(SUPERKERNEL_ARGS_CALL); |
| 18 | } | 19 | } |
| @@ -18,6 +18,7 @@ using namespace AscendC; | |||
| 18 | 18 | ||
| 19 | extern "C" __global__ __aicore__ void aiv_alltoall_# | 19 | extern "C" __global__ __aicore__ void aiv_alltoall_# |
| 20 | { \ | 20 | { \ |
| 21 | + AIV_INFO_HINT; \ | ||
| 21 | AivAlltoAllV2Mesh1D<type>(KERNEL_ARGS_CALL); \ | 22 | AivAlltoAllV2Mesh1D<type>(KERNEL_ARGS_CALL); \ |
| 22 | } \ | 23 | } \ |
| 23 | EXPORT_AIV_META_INFO(aiv_alltoall_# | 24 | EXPORT_AIV_META_INFO(aiv_alltoall_# |
| @@ -18,6 +18,7 @@ using namespace AscendC; | |||
| 18 | 18 | ||
| 19 | extern "C" __global__ __aicore__ void aiv_alltoallv_# | 19 | extern "C" __global__ __aicore__ void aiv_alltoallv_# |
| 20 | { \ | 20 | { \ |
| 21 | + AIV_INFO_HINT; \ | ||
| 21 | AivAlltoAllVV2Mesh1D<type>(EXTERN_KERNEL_ARGS_CALL); \ | 22 | AivAlltoAllVV2Mesh1D<type>(EXTERN_KERNEL_ARGS_CALL); \ |
| 22 | } \ | 23 | } \ |
| 23 | EXPORT_AIV_META_INFO(aiv_alltoallv_# | 24 | EXPORT_AIV_META_INFO(aiv_alltoallv_# |
| @@ -17,6 +17,7 @@ add_custom_target(hccl_aiv_broadcast_op_910_95 ALL | |||
| 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_95.o) | 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_95.o) |
| 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_95.o | 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_95.o |
| 19 | 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_95_tmp.o --cce-aicore-arch=dav-c310-vec | 19 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_broadcast_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_95_tmp.o --cce-aicore-arch=dav-c310-vec |
| 20 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 20 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 21 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 21 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 22 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| 22 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 23 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| @@ -41,6 +42,7 @@ add_custom_target(hccl_aiv_broadcast_op_910_96 ALL | |||
| 41 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_96.o) | 42 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_96.o) |
| 42 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_96.o | 43 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_96.o |
| 43 | 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 | 44 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_broadcast_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_broadcast_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec |
| 45 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 44 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 46 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 45 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ | 47 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ |
| 46 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw | 48 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw |
| @@ -19,6 +19,7 @@ using namespace AscendC; | |||
| 19 | 19 | ||
| 20 | extern "C" __global__ __aicore__ void aiv_broadcast_# | 20 | extern "C" __global__ __aicore__ void aiv_broadcast_# |
| 21 | { \ | 21 | { \ |
| 22 | + AIV_INFO_HINT; \ | ||
| 22 | return AivBroadcastV2Mesh1D<type>(KERNEL_ARGS_CALL); \ | 23 | return AivBroadcastV2Mesh1D<type>(KERNEL_ARGS_CALL); \ |
| 23 | } \ | 24 | } \ |
| 24 | EXPORT_AIV_META_INFO(aiv_broadcast_# | 25 | EXPORT_AIV_META_INFO(aiv_broadcast_# |
| @@ -156,4 +156,11 @@ enum class CommPattern { | |||
| 156 | 156 | ||
| 157 | 157 | ||
| 158 | 158 | ||
| 159 | + | ||
| 160 | + do { \ | ||
| 161 | + AscendC::PRINTF(format, ##__VA_ARGS__); \ | ||
| 162 | + } while (0) | ||
| 163 | + | ||
| 164 | + | ||
| 165 | + | ||
| 159 | 166 | ||
| @@ -17,6 +17,7 @@ add_custom_target(hccl_aiv_recv_op_910_95 ALL | |||
| 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_95.o) | 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_95.o) |
| 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_95.o | 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_95.o |
| 19 | 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_95_tmp.o --cce-aicore-arch=dav-c310-vec | 19 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_recv_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_95_tmp.o --cce-aicore-arch=dav-c310-vec |
| 20 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 20 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 21 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 21 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 22 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| 22 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 23 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| @@ -41,6 +42,7 @@ add_custom_target(hccl_aiv_recv_op_910_96 ALL | |||
| 41 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_96.o) | 42 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_96.o) |
| 42 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_96.o | 43 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_96.o |
| 43 | 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 | 44 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_recv_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_recv_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec |
| 45 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 44 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 46 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 45 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ | 47 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ |
| 46 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw | 48 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw |
| @@ -18,6 +18,7 @@ using namespace AscendC; | |||
| 18 | 18 | ||
| 19 | extern "C" __global__ __aicore__ void aiv_recv_# | 19 | extern "C" __global__ __aicore__ void aiv_recv_# |
| 20 | { \ | 20 | { \ |
| 21 | + AIV_INFO_HINT; \ | ||
| 21 | AivRecvV2Mesh1D<type>(KERNEL_ARGS_CALL); \ | 22 | AivRecvV2Mesh1D<type>(KERNEL_ARGS_CALL); \ |
| 22 | } \ | 23 | } \ |
| 23 | EXPORT_AIV_META_INFO(aiv_recv_# | 24 | EXPORT_AIV_META_INFO(aiv_recv_# |
| @@ -17,6 +17,7 @@ add_custom_target(hccl_aiv_reduce_op_910_95 ALL | |||
| 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_95.o) | 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_95.o) |
| 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_95.o | 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_95.o |
| 19 | 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_95_tmp.o --cce-aicore-arch=dav-c310-vec | 19 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_reduce_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_95_tmp.o --cce-aicore-arch=dav-c310-vec |
| 20 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 20 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 21 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 21 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 22 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| 22 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 23 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| @@ -41,6 +42,7 @@ add_custom_target(hccl_aiv_reduce_op_910_96 ALL | |||
| 41 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_96.o) | 42 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_96.o) |
| 42 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_96.o | 43 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_96.o |
| 43 | 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 | 44 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_reduce_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec |
| 45 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 44 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 46 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 45 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ | 47 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ |
| 46 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw | 48 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw |
| @@ -19,6 +19,7 @@ using namespace AscendC; | |||
| 19 | 19 | ||
| 20 | extern "C" __global__ __aicore__ void aiv_reduce_# | 20 | extern "C" __global__ __aicore__ void aiv_reduce_# |
| 21 | { \ | 21 | { \ |
| 22 | + AIV_INFO_HINT; \ | ||
| 22 | return AivReduceV2Mesh1D<type>(KERNEL_ARGS_CALL); \ | 23 | return AivReduceV2Mesh1D<type>(KERNEL_ARGS_CALL); \ |
| 23 | } \ | 24 | } \ |
| 24 | EXPORT_AIV_META_INFO(aiv_reduce_# | 25 | EXPORT_AIV_META_INFO(aiv_reduce_# |
| @@ -17,6 +17,7 @@ add_custom_target(hccl_aiv_reduce_scatter_op_910_95 ALL | |||
| 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_95.o) | 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_95.o) |
| 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_95.o | 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_95.o |
| 19 | 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_95_tmp.o --cce-aicore-arch=dav-c310-vec | 19 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_reduce_scatter_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_95_tmp.o --cce-aicore-arch=dav-c310-vec |
| 20 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 20 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 21 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 21 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 22 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| 22 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 23 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| @@ -41,7 +42,7 @@ add_custom_target(hccl_rs_superkernel_mesh_1d_int8_v2 ALL | |||
| 41 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int8_v2.o | 42 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int8_v2.o |
| 42 | 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.o | 43 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int8_v2.o |
| 43 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 44 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 44 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 45 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 45 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 46 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 46 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 47 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 47 | -DHCCL_DTYPE_INT8 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int8 | 48 | -DHCCL_DTYPE_INT8 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int8 |
| @@ -62,7 +63,7 @@ add_custom_target(hccl_rs_superkernel_mesh_1d_int16_v2 ALL | |||
| 62 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int16_v2.o | 63 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int16_v2.o |
| 63 | 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.o | 64 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int16_v2.o |
| 64 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 65 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 65 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 66 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 66 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 67 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 67 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 68 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 68 | -DHCCL_DTYPE_INT16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int16 | 69 | -DHCCL_DTYPE_INT16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int16 |
| @@ -83,7 +84,7 @@ add_custom_target(hccl_rs_superkernel_mesh_1d_int32_v2 ALL | |||
| 83 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int32_v2.o | 84 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int32_v2.o |
| 84 | 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.o | 85 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int32_v2.o |
| 85 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 86 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 86 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 87 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 87 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 88 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 88 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 89 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 89 | -DHCCL_DTYPE_INT32 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int32 | 90 | -DHCCL_DTYPE_INT32 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int32 |
| @@ -104,7 +105,7 @@ add_custom_target(hccl_rs_superkernel_mesh_1d_float16_v2 ALL | |||
| 104 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float16_v2.o | 105 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float16_v2.o |
| 105 | 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.o | 106 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float16_v2.o |
| 106 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 107 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 107 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 108 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 108 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 109 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 109 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 110 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 110 | -DHCCL_DTYPE_FP16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_float16 | 111 | -DHCCL_DTYPE_FP16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_float16 |
| @@ -125,7 +126,7 @@ add_custom_target(hccl_rs_superkernel_mesh_1d_float32_v2 ALL | |||
| 125 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float32_v2.o | 126 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float32_v2.o |
| 126 | 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.o | 127 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float32_v2.o |
| 127 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 128 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 128 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 129 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 129 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 130 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 130 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 131 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 131 | -DHCCL_DTYPE_FP32 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_float32 | 132 | -DHCCL_DTYPE_FP32 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_float32 |
| @@ -146,7 +147,7 @@ add_custom_target(hccl_rs_superkernel_mesh_1d_bfloat16_v2 ALL | |||
| 146 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_bfloat16_v2.o | 147 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_bfloat16_v2.o |
| 147 | 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.o | 148 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_bfloat16_v2.o |
| 148 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 149 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 149 | - -DASCENDC_DUMP --cce-aicore-arch=dav-c310-vec | 150 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-c310-vec |
| 150 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 151 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 151 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ | 152 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=310 -D__DAV_C310__ |
| 152 | -DHCCL_DTYPE_BFP16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_bfloat16 | 153 | -DHCCL_DTYPE_BFP16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_bfloat16 |
| @@ -203,6 +204,7 @@ add_custom_target(hccl_aiv_reduce_scatter_op_910_96 ALL | |||
| 203 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_96.o) | 204 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_96.o) |
| 204 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_96.o | 205 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_96.o |
| 205 | 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 | 206 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_reduce_scatter_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_reduce_scatter_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec |
| 207 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 206 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 208 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 207 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ | 209 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ |
| 208 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw | 210 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw |
| @@ -228,7 +230,7 @@ add_custom_target(hccl_rs_superkernel_mesh_1d_int8_v2_910_96 ALL | |||
| 228 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int8_v2_910_96.o | 230 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int8_v2_910_96.o |
| 229 | 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 | 231 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int8_v2_910_96.o |
| 230 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 232 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 231 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 233 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 232 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 234 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 233 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 235 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 234 | -DHCCL_DTYPE_INT8 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int8 | 236 | -DHCCL_DTYPE_INT8 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int8 |
| @@ -250,7 +252,7 @@ add_custom_target(hccl_rs_superkernel_mesh_1d_int16_v2_910_96 ALL | |||
| 250 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int16_v2_910_96.o | 252 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int16_v2_910_96.o |
| 251 | 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 | 253 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int16_v2_910_96.o |
| 252 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 254 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 253 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 255 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 254 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 256 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 255 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 257 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 256 | -DHCCL_DTYPE_INT16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int16 | 258 | -DHCCL_DTYPE_INT16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int16 |
| @@ -272,7 +274,7 @@ add_custom_target(hccl_rs_superkernel_mesh_1d_int32_v2_910_96 ALL | |||
| 272 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int32_v2_910_96.o | 274 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int32_v2_910_96.o |
| 273 | 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 | 275 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_int32_v2_910_96.o |
| 274 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 276 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 275 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 277 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 276 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 278 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 277 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 279 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 278 | -DHCCL_DTYPE_INT32 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int32 | 280 | -DHCCL_DTYPE_INT32 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_int32 |
| @@ -294,7 +296,7 @@ add_custom_target(hccl_rs_superkernel_mesh_1d_float16_v2_910_96 ALL | |||
| 294 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float16_v2_910_96.o | 296 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float16_v2_910_96.o |
| 295 | 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 | 297 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float16_v2_910_96.o |
| 296 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 298 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 297 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 299 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 298 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 300 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 299 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 301 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 300 | -DHCCL_DTYPE_FP16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_float16 | 302 | -DHCCL_DTYPE_FP16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_float16 |
| @@ -316,7 +318,7 @@ add_custom_target(hccl_rs_superkernel_mesh_1d_float32_v2_910_96 ALL | |||
| 316 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float32_v2_910_96.o | 318 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float32_v2_910_96.o |
| 317 | 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 | 319 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_float32_v2_910_96.o |
| 318 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 320 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 319 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 321 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 320 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 322 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 321 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 323 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 322 | -DHCCL_DTYPE_FP32 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_float32 | 324 | -DHCCL_DTYPE_FP32 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_float32 |
| @@ -338,7 +340,7 @@ add_custom_target(hccl_rs_superkernel_mesh_1d_bfloat16_v2_910_96 ALL | |||
| 338 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_bfloat16_v2_910_96.o | 340 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_bfloat16_v2_910_96.o |
| 339 | 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 | 341 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_rs_mesh_1d_superkernel.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_rs_superkernel_mesh_1d_bfloat16_v2_910_96.o |
| 340 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 | 342 | -DTILING_KEY_VAR=0UL -D__ASCENDC_ENABLE_SUPER_KERNEL__ -DL2_CACHE_HINT -DCONST_TILING -DDETERMINISTIC_MODE=0 |
| 341 | - -DASCENDC_DUMP --cce-aicore-arch=dav-920r1-vec | 343 | + -DASCENDC_DUMP=${ASCENDC_DUMP} --cce-aicore-arch=dav-920r1-vec |
| 342 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false | 344 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-dcci-insert-for-scalar=false |
| 343 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ | 345 | -mllvm -cce-aicore-stack-size=0x8000 -D__CCE_AICORE__=920 -D__DAV_920R1__ |
| 344 | -DHCCL_DTYPE_BFP16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_bfloat16 | 346 | -DHCCL_DTYPE_BFP16 -Dsk_reducescatter_mesh_1d=sk_reducescatter_mesh_1d_bfloat16 |
| @@ -23,6 +23,7 @@ using namespace AscendC; | |||
| 23 | 23 | ||
| 24 | extern "C" __global__ __aicore__ void aiv_reduce_scatter_# | 24 | extern "C" __global__ __aicore__ void aiv_reduce_scatter_# |
| 25 | { \ | 25 | { \ |
| 26 | + AIV_INFO_HINT; \ | ||
| 26 | if (numBlocks > 2 * rankSize) { \ | 27 | if (numBlocks > 2 * rankSize) { \ |
| 27 | AivReduceScatterV2Mesh1DBigData<type>(KERNEL_ARGS_CALL); \ | 28 | AivReduceScatterV2Mesh1DBigData<type>(KERNEL_ARGS_CALL); \ |
| 28 | } else if (numBlocks >= rankSize) { \ | 29 | } else if (numBlocks >= rankSize) { \ |
| @@ -48,6 +48,7 @@ __aicore__ inline void sk_rs_superkernel_dispatch(SUPERKERNEL_ARGS_DEF) | |||
| 48 | 48 | ||
| 49 | extern "C" __aicore__ void sk_reducescatter_mesh_1d(SUPERKERNEL_LITE_ARGS_DEF) | 49 | extern "C" __aicore__ void sk_reducescatter_mesh_1d(SUPERKERNEL_LITE_ARGS_DEF) |
| 50 | { | 50 | { |
| 51 | + AIV_INFO_HINT; | ||
| 51 | SUPERKERNEL_LITE_ARGS_EXTRACT; | 52 | SUPERKERNEL_LITE_ARGS_EXTRACT; |
| 52 | return sk_rs_superkernel_dispatch(SUPERKERNEL_ARGS_CALL); | 53 | return sk_rs_superkernel_dispatch(SUPERKERNEL_ARGS_CALL); |
| 53 | } | 54 | } |
| @@ -17,6 +17,7 @@ add_custom_target(hccl_aiv_scatter_op_910_95 ALL | |||
| 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_95.o) | 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_95.o) |
| 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_95.o | 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_95.o |
| 19 | 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_95_tmp.o --cce-aicore-arch=dav-c310-vec | 19 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_scatter_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_95_tmp.o --cce-aicore-arch=dav-c310-vec |
| 20 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 20 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 21 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 21 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 22 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| 22 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 23 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| @@ -41,6 +42,7 @@ add_custom_target(hccl_aiv_scatter_op_910_96 ALL | |||
| 41 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_96.o) | 42 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_96.o) |
| 42 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_96.o | 43 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_96.o |
| 43 | 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 | 44 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_scatter_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_scatter_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec |
| 45 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 44 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 46 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 45 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ | 47 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ |
| 46 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw | 48 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw |
| @@ -19,6 +19,7 @@ using namespace AscendC; | |||
| 19 | 19 | ||
| 20 | extern "C" __global__ __aicore__ void aiv_scatter_# | 20 | extern "C" __global__ __aicore__ void aiv_scatter_# |
| 21 | { \ | 21 | { \ |
| 22 | + AIV_INFO_HINT; \ | ||
| 22 | return AivScatterV2Mesh1D<type>(KERNEL_ARGS_CALL); \ | 23 | return AivScatterV2Mesh1D<type>(KERNEL_ARGS_CALL); \ |
| 23 | } \ | 24 | } \ |
| 24 | EXPORT_AIV_META_INFO(aiv_scatter_# | 25 | EXPORT_AIV_META_INFO(aiv_scatter_# |
| @@ -17,6 +17,7 @@ add_custom_target(hccl_aiv_send_op_910_95 ALL | |||
| 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_95.o) | 17 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_95.o) |
| 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_95.o | 18 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_95.o |
| 19 | 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_95_tmp.o --cce-aicore-arch=dav-c310-vec | 19 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_send_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_95_tmp.o --cce-aicore-arch=dav-c310-vec |
| 20 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 20 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 21 | --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 21 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 22 | "-I$<JOIN:$<TARGET_PROPERTY:asc_kernel_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| 22 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" | 23 | "-I$<JOIN:$<TARGET_PROPERTY:kernel_tiling_headers,INTERFACE_INCLUDE_DIRECTORIES>,;-I>" |
| @@ -41,6 +42,7 @@ add_custom_target(hccl_aiv_send_op_910_96 ALL | |||
| 41 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_96.o) | 42 | DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_96.o) |
| 42 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_96.o | 43 | add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_96.o |
| 43 | 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 | 44 | COMMAND ${CCE_DIR}/ccec -c -x cce -O2 ${CMAKE_CURRENT_SOURCE_DIR}/aiv_send_op.h -o ${CMAKE_CURRENT_BINARY_DIR}/hccl_aiv_send_op_910_96_tmp.o --cce-aicore-arch=dav-920r1-vec |
| 45 | + -DASCENDC_DUMP=${ASCENDC_DUMP} | ||
| 44 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 | 46 | --cce-aicore-input-parameter-size=28000 --cce-aicore-only -mllvm -cce-aicore-function-stack-size=0x8000 -mllvm -cce-aicore-stack-size=0x8000 |
| 45 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ | 47 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw/lib/ |
| 46 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw | 48 | -I${ASCEND_CANN_PACKAGE_PATH}/tools/tikcpp/tikcfw |
| @@ -18,6 +18,7 @@ using namespace AscendC; | |||
| 18 | 18 | ||
| 19 | extern "C" __global__ __aicore__ void aiv_send_# | 19 | extern "C" __global__ __aicore__ void aiv_send_# |
| 20 | { \ | 20 | { \ |
| 21 | + AIV_INFO_HINT; \ | ||
| 21 | AivSendV2Mesh1D<type>(KERNEL_ARGS_CALL); \ | 22 | AivSendV2Mesh1D<type>(KERNEL_ARGS_CALL); \ |
| 22 | } \ | 23 | } \ |
| 23 | EXPORT_AIV_META_INFO(aiv_send_# | 24 | EXPORT_AIV_META_INFO(aiv_send_# |