kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-22 19:52:38 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

remove sm70 from fmha_v2 completely

Browse Source 
Signed-off-by: Xiwen Yu <13230610+VALLIS-NERIA@users.noreply.github.com>
This commit is contained in:
Xiwen Yu 2025-09-12 23:08:51 +08:00
parent 98cbab0b34
commit 514ebc203e
1 changed files with 0 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
cpp/kernels/fmha_v2/Makefile
View File

@ -90,7 +90,6 @@ NVCC_FLAGS += $(PREPROCESSOR_FLAGS)
# The include directories.
INCLUDE_DIRS += -I./src -I./generated -I$(CUDA)/include
GENCODE_SM70 = -gencode=arch=compute_70,code=\"sm_70\"
GENCODE_SM72 = -gencode=arch=compute_72,code=\"sm_72\"
GENCODE_SM75 = -gencode=arch=compute_75,code=\"sm_75\"
GENCODE_SM80 = -gencode=arch=compute_80,code=\"sm_80\"
@ -127,10 +126,6 @@ CUBIN_OBJ = $(patsubst %.cubin.cpp, %.cubin.o, $(CUBIN_CPP))
GENCODES =
ifdef ENABLE_SM70
GENCODES += $(GENCODE_SM70)
endif
GENCODES += $(GENCODE_SM72)
GENCODES += $(GENCODE_SM75)
GENCODES += $(GENCODE_SM80)
@ -157,20 +152,12 @@ UNIT_TEST_OBJ = $(patsubst %.cu, obj/%.o, $(UNIT_TEST_CPP))
UNIT_TEST_EXE = $(patsubst %.cu, bin/%.exe, $(UNIT_TEST_CPP))
# arch-dependent boilerplates
UNIT_TEST_CPP_SM70 =
ifdef ENABLE_SM70
UNIT_TEST_CPP_SM70 = $(wildcard $(UNIT_TEST_CPP_DIR)/arch/*_sm70.cu)
UNIT_TEST_OBJ_SM70 = $(patsubst %_sm70.cu, obj/%_sm70.o, $(UNIT_TEST_CPP_SM70))
UNIT_TEST_EXE_SM70 = $(patsubst %_sm70.cu, bin/%_sm70.exe, $(UNIT_TEST_CPP_SM70))
endif
UNIT_TEST_CPP_SM80 = $(wildcard $(UNIT_TEST_CPP_DIR)/arch/*_sm80.cu)
UNIT_TEST_OBJ_SM80 = $(patsubst %_sm80.cu, obj/%_sm80.o, $(UNIT_TEST_CPP_SM80))
UNIT_TEST_EXE_SM80 = $(patsubst %_sm80.cu, bin/%_sm80.exe, $(UNIT_TEST_CPP_SM80))
# aggregate exes as prerequisite of build target "test"
UNIT_TEST_EXE_ARCH =
UNIT_TEST_EXE_ARCH += $(UNIT_TEST_EXE_SM70)
UNIT_TEST_EXE_ARCH += $(UNIT_TEST_EXE_SM80)
# #################################################################################################
@ -253,8 +240,6 @@ bin/libfmha_cubin.a: $(CUBIN_OBJ)
###################################################################################################
obj/%_sm70.cu.o: ./generated/%_sm70.cu ./src/*.h ./src/fmha/*.h
$(NVCC) $(NVCC_FLAGS) $(I2F_F2I_FLAGS) $(GENCODE_SM70) $(INCLUDE_DIRS) -c -o $@ $<
obj/%_sm72.cu.o: ./generated/%_sm72.cu ./src/*.h ./src/fmha/*.h
$(NVCC) $(NVCC_FLAGS) $(I2F_F2I_FLAGS) $(GENCODE_SM72) $(INCLUDE_DIRS) -c -o $@ $<
obj/%_sm75.cu.o: ./generated/%_sm75.cu ./src/*.h ./src/fmha/*.h
@ -274,8 +259,6 @@ obj/%_sm100.cu.o: ./generated/%_sm100.cu ./src/*.h ./src/fmha/*.h ./src/fmha/hop
obj/%_sm120.cu.o: ./generated/%_sm120.cu ./src/*.h ./src/fmha/*.h
$(NVCC) $(NVCC_FLAGS) $(I2F_F2I_FLAGS) $(GENCODE_SM120) $(INCLUDE_DIRS) -c -o $@ $<
obj/%_sm70.no_i2f_f2i.cu.o: ./generated/%_sm70.no_i2f_f2i.cu ./src/*.h ./src/fmha/*.h
$(NVCC) $(NVCC_FLAGS) $(GENCODE_SM70) $(INCLUDE_DIRS) -c -o $@ $<
obj/%_sm72.no_i2f_f2i.cu.o: ./generated/%_sm72.no_i2f_f2i.cu ./src/*.h ./src/fmha/*.h
$(NVCC) $(NVCC_FLAGS) $(GENCODE_SM72) $(INCLUDE_DIRS) -c -o $@ $<
obj/%_sm75.no_i2f_f2i.cu.o: ./generated/%_sm75.no_i2f_f2i.cu ./src/*.h ./src/fmha/*.h
@ -319,16 +302,11 @@ $(UNIT_TEST_OBJ): $(UNIT_TEST_OBJ_DIR)/%.o : ${UNIT_TEST_CPP_DIR}/%.cu ./src/*.h
$(NVCC) $(NVCC_FLAGS) $(I2F_F2I_FLAGS) $(GENCODES) -c -o $@ $< -I./src $(GTEST_INC)
# arch-dependent objs
$(UNIT_TEST_OBJ_SM70): %.o : $(UNIT_TEST_CPP_SM70) ./src/*.h ./src/fmha/*.h
$(NVCC) $(NVCC_FLAGS) $(I2F_F2I_FLAGS) $(GENCODE_SM70) -c -o $@ $< -I./src $(GTEST_INC)
$(UNIT_TEST_OBJ_SM80): %.o : $(UNIT_TEST_CPP_SM80) ./src/*.h ./src/fmha/*.h
$(NVCC) $(NVCC_FLAGS) $(I2F_F2I_FLAGS) $(GENCODE_SM80) -c -o $@ $< -I./src $(GTEST_INC)
###################################################################################################
cubin/%_sm70.cu.cubin: ./generated/%_sm70.cu ./src/*.h ./src/fmha/*.h
$(NVCC) $(NVCC_FLAGS) $(I2F_F2I_FLAGS) $(GENCODE_SM70) $(INCLUDE_DIRS) -cubin -o $@ $<
cubin/%_sm72.cu.cubin: ./generated/%_sm72.cu ./src/*.h ./src/fmha/*.h
$(NVCC) $(NVCC_FLAGS) $(I2F_F2I_FLAGS) $(GENCODE_SM72) $(INCLUDE_DIRS) -cubin -o $@ $<
cubin/%_sm75.cu.cubin: ./generated/%_sm75.cu ./src/*.h ./src/fmha/*.h
@ -348,8 +326,6 @@ cubin/%_sm100.cu.cubin: ./generated/%_sm100.cu ./src/*.h ./src/fmha/*.h
cubin/%_sm120.cu.cubin: ./generated/%_sm120.cu ./src/*.h ./src/fmha/*.h
$(NVCC) $(NVCC_FLAGS) $(I2F_F2I_FLAGS) $(GENCODE_SM120) $(INCLUDE_DIRS) -cubin -o $@ $<
cubin/%_sm70.no_i2f_f2i.cu.cubin: ./generated/%_sm70.no_i2f_f2i.cu ./src/*.h ./src/fmha/*.h
$(NVCC) $(NVCC_FLAGS) $(GENCODE_SM70) $(INCLUDE_DIRS) -cubin -o $@ $<
cubin/%_sm72.no_i2f_f2i.cu.cubin: ./generated/%_sm72.no_i2f_f2i.cu ./src/*.h ./src/fmha/*.h
$(NVCC) $(NVCC_FLAGS) $(GENCODE_SM72) $(INCLUDE_DIRS) -cubin -o $@ $<
cubin/%_sm75.no_i2f_f2i.cu.cubin: ./generated/%_sm75.no_i2f_f2i.cu ./src/*.h ./src/fmha/*.h