You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

Makefile 2.1 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657
  1. PARAM_DEFS := ../src/common/elemwise/each_mode.inl
  2. ELEMWISE_IMPL := ../src/cuda/cond_take/kimpl \
  3. ../src/cuda/elemwise/special_kimpl \
  4. ../src/cuda/elemwise/kimpl \
  5. ../src/rocm/elemwise/special_kimpl \
  6. ../src/rocm/elemwise/kimpl \
  7. ../src/naive/elemwise/kimpl \
  8. ../src/cuda/elemwise_multi_type/kimpl
  9. CUDA_CONV_IMPL := ../src/cuda/conv_bias/int8/kimpl ../src/cuda/conv_bias/int8_imma/kimpl ../src/cuda/batch_conv_bias/int8/kimpl
  10. CUDA_MATMUL_IMPL := ../src/cuda/matrix_mul/fp32_simt/kimpl ../src/cuda/matrix_mul/fp32_simt_gemv/kimpl
  11. all: ${PARAM_DEFS} ${ELEMWISE_IMPL} ${CUDA_CONV_IMPL} $(CUDA_MATMUL_IMPL)
  12. ../src/common/elemwise/each_mode.inl: gen_elemwise_each_mode.py
  13. ./$^ $@
  14. ../src/cuda/cond_take/kimpl: gen_cond_take_kern_impls.py
  15. ./$^ --type cuda $@
  16. ../src/cuda/elemwise/special_kimpl: gen_elemwise_special_kern_impls.py
  17. ./$^ --type cuda $@
  18. ../src/rocm/elemwise/special_kimpl: gen_elemwise_special_kern_impls.py
  19. ./$^ --type hip $@
  20. ../src/cuda/elemwise/kimpl: gen_elemwise_kern_impls.py
  21. ./$^ --type cuda $@
  22. ../src/rocm/elemwise/kimpl: gen_elemwise_kern_impls.py
  23. ./$^ --type hip $@
  24. ../src/%/elemwise/kimpl: gen_elemwise_kern_impls.py
  25. ./$^ $@
  26. ../src/cuda/elemwise_multi_type/kimpl: gen_elemwise_multi_type_kern_impls.py
  27. ./$^ --type cuda $@
  28. ../src/cuda/conv_bias/int8/kimpl: gen_cuda_conv_bias_kern_impls.py cutlass_generator/generator.py
  29. ./gen_cuda_conv_bias_kern_impls.py --type dp4a $@
  30. python3 ./cutlass_generator/generator.py --operations conv2d --type simt $@
  31. ../src/cuda/conv_bias/int8_imma/kimpl: gen_cuda_conv_bias_kern_impls.py cutlass_generator/generator.py
  32. ./gen_cuda_conv_bias_kern_impls.py --type imma $@
  33. python3 ./cutlass_generator/generator.py --operations conv2d --type tensorop8816 $@
  34. ../src/cuda/batch_conv_bias/int8/kimpl: gen_cuda_batch_conv_bias_kern_impls.py
  35. ./$^ --type dp4a $@
  36. ../src/cuda/matrix_mul/fp32_simt/kimpl: cutlass_generator/generator.py
  37. python3 ./cutlass_generator/generator.py --operations gemm --type simt $@
  38. ../src/cuda/matrix_mul/fp32_simt_gemv/kimpl: cutlass_generator
  39. python3 ./cutlass_generator/generator.py --operations gemv --type simt $@
  40. .PHONY: all