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 6.1 kB

6 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207
  1. ifdef TARGET_CORE
  2. TARGET = $(TARGET_CORE)
  3. endif
  4. TOPDIR = ..
  5. include $(TOPDIR)/Makefile.system
  6. AVX2OPT =
  7. ifeq ($(C_COMPILER), GCC)
  8. # AVX2 support was added in 4.7.0
  9. GCCVERSIONCHECK := $(GCCVERSIONGT4)$(GCCVERSIONGTEQ4)$(GCCMINORVERSIONGTEQ7)
  10. ifeq ($(GCCVERSIONCHECK), $(filter $(GCCVERSIONCHECK), 011 110 111))
  11. AVX2OPT = -mavx2
  12. endif
  13. endif
  14. ifeq ($(C_COMPILER), CLANG)
  15. # Any clang posing as gcc 4.2 should be new enough (3.4 or later)
  16. GCCVERSIONCHECK := $(GCCVERSIONGT4)$(GCCVERSIONGTEQ4)$(GCCMINORVERSIONGTEQ2)
  17. ifeq ($(GCCVERSIONCHECK), $(filter $(GCCVERSIONCHECK), 011 110 111))
  18. AVX2OPT = -mavx2 -mfma
  19. endif
  20. endif
  21. ifdef NO_AVX2
  22. AVX2OPT=
  23. endif
  24. ifdef TARGET_CORE
  25. ifeq ($(TARGET_CORE), SAPPHIRERAPIDS)
  26. override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
  27. ifeq (1, $(filter 1,$(GCCVERSIONGTEQ11) $(CLANGVERSIONGTEQ12)))
  28. override CFLAGS += -march=sapphirerapids
  29. else
  30. override CFLAGS += -march=skylake-avx512 -mavx512f
  31. endif
  32. ifeq ($(OSNAME), CYGWIN_NT)
  33. override CFLAGS += -fno-asynchronous-unwind-tables
  34. endif
  35. ifeq ($(OSNAME), WINNT)
  36. ifeq ($(C_COMPILER), GCC)
  37. override CFLAGS += -fno-asynchronous-unwind-tables
  38. endif
  39. endif
  40. else ifeq ($(TARGET_CORE), COOPERLAKE)
  41. override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
  42. ifeq (1, $(filter 1,$(GCCVERSIONGTEQ10) $(CLANGVERSIONGTEQ9)))
  43. override CFLAGS += -march=cooperlake
  44. else
  45. override CFLAGS += -march=skylake-avx512 -mavx512f
  46. endif
  47. ifeq ($(OSNAME), CYGWIN_NT)
  48. override CFLAGS += -fno-asynchronous-unwind-tables
  49. endif
  50. ifeq ($(OSNAME), WINNT)
  51. ifeq ($(C_COMPILER), GCC)
  52. override CFLAGS += -fno-asynchronous-unwind-tables
  53. endif
  54. endif
  55. else ifeq ($(TARGET_CORE), SKYLAKEX)
  56. override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) -march=skylake-avx512 -mavx512f
  57. ifeq ($(OSNAME), CYGWIN_NT)
  58. override CFLAGS += -fno-asynchronous-unwind-tables
  59. endif
  60. ifeq ($(OSNAME), WINNT)
  61. ifeq ($(C_COMPILER), GCC)
  62. override CFLAGS += -fno-asynchronous-unwind-tables
  63. endif
  64. endif
  65. else ifeq ($(TARGET_CORE), HASWELL)
  66. override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) $(AVX2OPT)
  67. else ifeq ($(TARGET_CORE), ZEN)
  68. override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) $(AVX2OPT)
  69. else ifeq ($(TARGET_CORE), LOONGSON3R4)
  70. override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) $(MSA_FLAGS)
  71. else ifneq ($(filter NEOVERSEN2 NEOVERSEV1, $(TARGET_CORE)),)
  72. ifeq ($(C_COMPILER), PGI)
  73. override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) -Msve_intrinsics
  74. else
  75. override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
  76. endif
  77. else
  78. override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
  79. endif
  80. BUILD_KERNEL = 1
  81. KDIR =
  82. TSUFFIX = _$(TARGET_CORE)
  83. else
  84. TARGET_CORE = $(CORE)
  85. KDIR =
  86. TSUFFIX =
  87. ifeq ($(TARGET_CORE), LOONGSON3R4)
  88. override CFLAGS += $(MSA_FLAGS)
  89. endif
  90. endif
  91. -include $(KERNELDIR)/KERNEL.$(TARGET_CORE)
  92. include $(KERNELDIR)/KERNEL
  93. include Makefile.L1
  94. include Makefile.L2
  95. include Makefile.L3
  96. include Makefile.LA
  97. HPLOBJS = \
  98. dgemm_kernel.$(SUFFIX) \
  99. $(DGEMMINCOPYOBJ) $(DGEMMITCOPYOBJ) \
  100. $(DGEMMONCOPYOBJ) $(DGEMMOTCOPYOBJ) \
  101. dtrsm_kernel_LN.$(SUFFIX) dtrsm_kernel_LT.$(SUFFIX) \
  102. dtrsm_kernel_RN.$(SUFFIX) dtrsm_kernel_RT.$(SUFFIX) \
  103. daxpy_k.$(SUFFIX) dcopy_k.$(SUFFIX) ddot_k.$(SUFFIX) \
  104. dger_k.$(SUFFIX) dscal_k.$(SUFFIX) idamax_k.$(SUFFIX) \
  105. dgemv_n.$(SUFFIX) dgemv_t.$(SUFFIX) dgemm_beta.$(SUFFIX) \
  106. dtrsm_iunucopy.$(SUFFIX) dtrsm_iunncopy.$(SUFFIX) \
  107. dtrsm_ilnucopy.$(SUFFIX) dtrsm_ilnncopy.$(SUFFIX) \
  108. dtrsm_iutucopy.$(SUFFIX) dtrsm_iutncopy.$(SUFFIX) \
  109. dtrsm_iltucopy.$(SUFFIX) dtrsm_iltncopy.$(SUFFIX) \
  110. dtrsm_ounucopy.$(SUFFIX) dtrsm_ounncopy.$(SUFFIX) \
  111. dtrsm_olnucopy.$(SUFFIX) dtrsm_olnncopy.$(SUFFIX) \
  112. dtrsm_outucopy.$(SUFFIX) dtrsm_outncopy.$(SUFFIX) \
  113. dtrsm_oltucopy.$(SUFFIX) dtrsm_oltncopy.$(SUFFIX)
  114. COMMONOBJS += lsame.$(SUFFIX) scabs1.$(SUFFIX) dcabs1.$(SUFFIX)
  115. ifeq ($(DYNAMIC_ARCH), 1)
  116. SBLASOBJS += setparam$(TSUFFIX).$(SUFFIX)
  117. CCOMMON_OPT += -DTS=$(TSUFFIX)
  118. endif
  119. KERNEL_INTERFACE = ../common_level1.h ../common_level2.h ../common_level3.h
  120. ifneq ($(NO_LAPACK), 1)
  121. KERNEL_INTERFACE += ../common_lapack.h
  122. endif
  123. ifeq ($(ARCH), x86)
  124. COMMONOBJS += cpuid.$(SUFFIX)
  125. endif
  126. ifdef EXPRECISION
  127. COMMONOBJS += qconjg.$(SUFFIX) qcabs1.$(SUFFIX)
  128. endif
  129. ifdef QUAD_PRECISION
  130. COMMONOBJS += qconjg.$(SUFFIX) qcabs1.$(SUFFIX)
  131. endif
  132. all : libs
  133. scabs1.$(SUFFIX): $(KERNELDIR)/$(SCABS_KERNEL)
  134. $(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE -DF_INTERFACE $< -o $(@F)
  135. dcabs1.$(SUFFIX): $(KERNELDIR)/$(DCABS_KERNEL)
  136. $(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -DF_INTERFACE $< -o $(@F)
  137. qcabs1.$(SUFFIX): $(KERNELDIR)/$(QCABS_KERNEL)
  138. $(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DF_INTERFACE $< -o $(@F)
  139. qconjg.$(SUFFIX): $(KERNELDIR)/qconjg.S
  140. $(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DF_INTERFACE $< -o $(@F)
  141. lsame.$(SUFFIX): $(KERNELDIR)/$(LSAME_KERNEL)
  142. $(CC) -c $(CFLAGS) -DF_INTERFACE $< -o $(@F)
  143. setparam$(TSUFFIX).$(SUFFIX): setparam$(TSUFFIX).c kernel$(TSUFFIX).h
  144. ifeq ($(USE_GEMM3M), 1)
  145. $(CC) -c $(CFLAGS) -DUSE_GEMM3M $< -o $@
  146. else
  147. $(CC) -c $(CFLAGS) $< -o $@
  148. endif
  149. setparam$(TSUFFIX).c : setparam-ref.c
  150. sed 's/TS/$(TSUFFIX)/g' $< > $(@F)
  151. kernel$(TSUFFIX).h : $(KERNEL_INTERFACE)
  152. sed 's/\ *(/$(TSUFFIX)(/g' $^ > $(@F)
  153. cpuid.$(SUFFIX): $(KERNELDIR)/cpuid.S
  154. $(CC) -c $(CFLAGS) $< -o $(@F)
  155. scabs1.$(PSUFFIX): $(KERNELDIR)/$(SCABS_KERNEL)
  156. $(CC) -c $(PFLAGS) -DCOMPLEX -UDOUBLE -DF_INTERFACE $< -o $(@F)
  157. dcabs1.$(PSUFFIX): $(KERNELDIR)/$(DCABS_KERNEL)
  158. $(CC) -c $(PFLAGS) -DCOMPLEX -DDOUBLE -DF_INTERFACE $< -o $(@F)
  159. qcabs1.$(PSUFFIX): $(KERNELDIR)/$(QCABS_KERNEL)
  160. $(CC) -c $(PFLAGS) -DCOMPLEX -DXDOUBLE -DF_INTERFACE $< -o $(@F)
  161. qconjg.$(PSUFFIX): $(KERNELDIR)/qconjg.S
  162. $(CC) -c $(PFLAGS) -DCOMPLEX -DXDOUBLE -DF_INTERFACE $< -o $(@F)
  163. lsame.$(PSUFFIX): $(KERNELDIR)/$(LSAME_KERNEL)
  164. $(CC) -c $(PFLAGS) -DF_INTERFACE $< -o $(@F)
  165. cpuid.$(PSUFFIX): $(KERNELDIR)/cpuid.S
  166. $(CC) -c $(PFLAGS) $< -o $(@F)
  167. #ifdef DYNAMIC_ARCH
  168. clean ::
  169. @rm -f setparam_*.c kernel_*.h setparam.h kernel.h
  170. #endif
  171. include $(TOPDIR)/Makefile.tail