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.system 34 kB

7 years ago
4 years ago
13 years ago
5 years ago
12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715
  1. #
  2. # Include user definition
  3. #
  4. # TO suppress recursive includes
  5. INCLUDED = 1
  6. ifndef TOPDIR
  7. TOPDIR = .
  8. endif
  9. # If ARCH is not set, we use the host system's architecture for getarch compile options.
  10. ifndef ARCH
  11. HOSTARCH := $(shell uname -m)
  12. else
  13. HOSTARCH = $(ARCH)
  14. endif
  15. HAVE_GAS := $(shell as -v < /dev/null 2>&1 | grep GNU 2>&1 >/dev/null)
  16. # Catch conflicting usage of ARCH in some BSD environments
  17. ifeq ($(ARCH), amd64)
  18. override ARCH=x86_64
  19. else ifeq ($(ARCH), powerpc64)
  20. override ARCH=power
  21. else ifeq ($(ARCH), powerpc64le)
  22. override ARCH=power
  23. else ifeq ($(ARCH), powerpc)
  24. override ARCH=power
  25. else ifeq ($(ARCH), i386)
  26. override ARCH=x86
  27. else ifeq ($(ARCH), armv6)
  28. override ARCH=arm
  29. else ifeq ($(ARCH), armv7)
  30. override ARCH=arm
  31. else ifeq ($(ARCH), aarch64)
  32. override ARCH=arm64
  33. else ifeq ($(ARCH), mipsel)
  34. override ARCH=mips
  35. else ifeq ($(ARCH), mips64el)
  36. override ARCH=mips64
  37. else ifeq ($(ARCH), zarch)
  38. override ARCH=zarch
  39. endif
  40. NETLIB_LAPACK_DIR = $(TOPDIR)/lapack-netlib
  41. # Default C compiler
  42. # - Only set if not specified on the command line or inherited from the environment.
  43. # - CC is an implicit variable so neither '?=' or 'ifndef' can be used.
  44. # http://stackoverflow.com/questions/4029274/mingw-and-make-variables
  45. # - Default value is 'cc' which is not always a valid command (e.g. MinGW).
  46. ifeq ($(origin CC),default)
  47. # Check if $(CC) refers to a valid command and set the value to gcc if not
  48. ifneq ($(findstring cmd.exe,$(SHELL)),)
  49. ifeq ($(shell where $(CC) 2>NUL),)
  50. CC = gcc
  51. endif
  52. else # POSIX-ish
  53. ifeq ($(shell command -v $(CC) 2>/dev/null),)
  54. ifeq ($(shell uname -s),Darwin)
  55. CC = clang
  56. # EXTRALIB += -Wl,-no_compact_unwind
  57. else
  58. CC = gcc
  59. endif # Darwin
  60. endif # CC exists
  61. endif # Shell is sane
  62. endif # CC is set to default
  63. # Default Fortran compiler (FC) is selected by f_check.
  64. ifndef MAKEFILE_RULE
  65. include $(TOPDIR)/Makefile.rule
  66. else
  67. include $(TOPDIR)/$(MAKEFILE_RULE)
  68. endif
  69. #
  70. # Beginning of system configuration
  71. #
  72. ifneq ($(BUILD_SINGLE),1)
  73. ifneq ($(BUILD_DOUBLE),1)
  74. ifneq ($(BUILD_COMPLEX),1)
  75. ifneq ($(BUILD_COMPLEX16),1)
  76. override BUILD_SINGLE=1
  77. override BUILD_DOUBLE=1
  78. override BUILD_COMPLEX=1
  79. override BUILD_COMPLEX16=1
  80. endif
  81. endif
  82. endif
  83. endif
  84. ifndef HOSTCC
  85. HOSTCC = $(CC)
  86. endif
  87. ifdef TARGET
  88. GETARCH_FLAGS := -DFORCE_$(TARGET)
  89. GETARCH_FLAGS += -DUSER_TARGET
  90. ifeq ($(TARGET), GENERIC)
  91. ifeq ($(DYNAMIC_ARCH), 1)
  92. override NO_EXPRECISION=1
  93. export NO_EXPRECiSION
  94. endif
  95. endif
  96. endif
  97. # Force fallbacks for 32bit
  98. ifeq ($(BINARY), 32)
  99. ifeq ($(TARGET), HASWELL)
  100. GETARCH_FLAGS := -DFORCE_NEHALEM
  101. endif
  102. ifeq ($(TARGET), SKYLAKEX)
  103. GETARCH_FLAGS := -DFORCE_NEHALEM
  104. endif
  105. ifeq ($(TARGET), COOPERLAKE)
  106. GETARCH_FLAGS := -DFORCE_NEHALEM
  107. endif
  108. ifeq ($(TARGET), SANDYBRIDGE)
  109. GETARCH_FLAGS := -DFORCE_NEHALEM
  110. endif
  111. ifeq ($(TARGET), BULLDOZER)
  112. GETARCH_FLAGS := -DFORCE_BARCELONA
  113. endif
  114. ifeq ($(TARGET), PILEDRIVER)
  115. GETARCH_FLAGS := -DFORCE_BARCELONA
  116. endif
  117. ifeq ($(TARGET), STEAMROLLER)
  118. GETARCH_FLAGS := -DFORCE_BARCELONA
  119. endif
  120. ifeq ($(TARGET), EXCAVATOR)
  121. GETARCH_FLAGS := -DFORCE_BARCELONA
  122. endif
  123. ifeq ($(TARGET), ZEN)
  124. GETARCH_FLAGS := -DFORCE_BARCELONA
  125. endif
  126. ifeq ($(TARGET), ARMV8)
  127. GETARCH_FLAGS := -DFORCE_ARMV7
  128. endif
  129. ifeq ($(TARGET), POWER8)
  130. GETARCH_FLAGS := -DFORCE_POWER6
  131. endif
  132. endif
  133. #TARGET_CORE will override TARGET which is used in DYNAMIC_ARCH=1.
  134. #
  135. ifdef TARGET_CORE
  136. GETARCH_FLAGS := -DFORCE_$(TARGET_CORE)
  137. endif
  138. # Force fallbacks for 32bit
  139. ifeq ($(BINARY), 32)
  140. ifeq ($(TARGET_CORE), HASWELL)
  141. GETARCH_FLAGS := -DFORCE_NEHALEM
  142. endif
  143. ifeq ($(TARGET_CORE), SKYLAKEX)
  144. GETARCH_FLAGS := -DFORCE_NEHALEM
  145. endif
  146. ifeq ($(TARGET_CORE), COOPERLAKE)
  147. GETARCH_FLAGS := -DFORCE_NEHALEM
  148. endif
  149. ifeq ($(TARGET_CORE), SANDYBRIDGE)
  150. GETARCH_FLAGS := -DFORCE_NEHALEM
  151. endif
  152. ifeq ($(TARGET_CORE), BULLDOZER)
  153. GETARCH_FLAGS := -DFORCE_BARCELONA
  154. endif
  155. ifeq ($(TARGET_CORE), PILEDRIVER)
  156. GETARCH_FLAGS := -DFORCE_BARCELONA
  157. endif
  158. ifeq ($(TARGET_CORE), STEAMROLLER)
  159. GETARCH_FLAGS := -DFORCE_BARCELONA
  160. endif
  161. ifeq ($(TARGET_CORE), EXCAVATOR)
  162. GETARCH_FLAGS := -DFORCE_BARCELONA
  163. endif
  164. ifeq ($(TARGET_CORE), ZEN)
  165. GETARCH_FLAGS := -DFORCE_BARCELONA
  166. endif
  167. endif
  168. # On x86_64 build getarch with march=native unless the compiler is PGI. This is required to detect AVX512 support in getarch.
  169. ifeq ($(HOSTARCH), x86_64)
  170. ifeq ($(findstring pgcc,$(HOSTCC))$(findstring nvc,$(HOSTCC)),)
  171. GETARCH_FLAGS += -march=native
  172. endif
  173. endif
  174. ifdef INTERFACE64
  175. ifneq ($(INTERFACE64), 0)
  176. GETARCH_FLAGS += -DUSE64BITINT
  177. endif
  178. endif
  179. ifndef GEMM_MULTITHREAD_THRESHOLD
  180. GEMM_MULTITHREAD_THRESHOLD=4
  181. endif
  182. GETARCH_FLAGS += -DGEMM_MULTITHREAD_THRESHOLD=$(GEMM_MULTITHREAD_THRESHOLD)
  183. ifeq ($(NO_AVX), 1)
  184. GETARCH_FLAGS += -DNO_AVX
  185. endif
  186. ifeq ($(BINARY), 32)
  187. GETARCH_FLAGS += -DNO_AVX -DNO_AVX2 -DNO_AVX512
  188. NO_AVX512 = 1
  189. endif
  190. ifeq ($(NO_AVX2), 1)
  191. GETARCH_FLAGS += -DNO_AVX2
  192. endif
  193. ifeq ($(NO_AVX512), 1)
  194. GETARCH_FLAGS += -DNO_AVX512
  195. endif
  196. ifeq ($(DEBUG), 1)
  197. GETARCH_FLAGS += -g
  198. endif
  199. ifeq ($(QUIET_MAKE), 1)
  200. MAKE += -s
  201. endif
  202. ifndef NO_PARALLEL_MAKE
  203. NO_PARALLEL_MAKE=0
  204. endif
  205. GETARCH_FLAGS += -DNO_PARALLEL_MAKE=$(NO_PARALLEL_MAKE)
  206. ifdef MAKE_NB_JOBS
  207. GETARCH_FLAGS += -DMAKE_NB_JOBS=$(MAKE_NB_JOBS)
  208. endif
  209. ifeq ($(HOSTCC), loongcc)
  210. GETARCH_FLAGS += -static
  211. endif
  212. #if don't use Fortran, it will only compile CBLAS.
  213. ifeq ($(ONLY_CBLAS), 1)
  214. NO_LAPACK = 1
  215. else
  216. ONLY_CBLAS = 0
  217. endif
  218. #For small matrix optimization
  219. ifeq ($(ARCH), x86_64)
  220. SMALL_MATRIX_OPT = 1
  221. endif
  222. ifeq ($(SMALL_MATRIX_OPT), 1)
  223. CCOMMON_OPT += -DSMALL_MATRIX_OPT
  224. endif
  225. # This operation is expensive, so execution should be once.
  226. ifndef GOTOBLAS_MAKEFILE
  227. export GOTOBLAS_MAKEFILE = 1
  228. # Generating Makefile.conf and config.h
  229. DUMMY := $(shell $(MAKE) -C $(TOPDIR) -f Makefile.prebuild CC="$(CC)" FC="$(FC)" HOSTCC="$(HOSTCC)" HOST_CFLAGS="$(GETARCH_FLAGS)" CFLAGS="$(CFLAGS)" BINARY=$(BINARY) USE_OPENMP=$(USE_OPENMP) TARGET_CORE=$(TARGET_CORE) ONLY_CBLAS=$(ONLY_CBLAS) TARGET=$(TARGET) all)
  230. ifndef TARGET_CORE
  231. include $(TOPDIR)/Makefile.conf
  232. else
  233. HAVE_NEON=
  234. HAVE_VFP=
  235. HAVE_VFPV3=
  236. HAVE_VFPV4=
  237. HAVE_MMX=
  238. HAVE_SSE=
  239. HAVE_SSE2=
  240. HAVE_SSE3=
  241. HAVE_SSSE3=
  242. HAVE_SSE4_1=
  243. HAVE_SSE4_2=
  244. HAVE_SSE4A=
  245. HAVE_SSE5=
  246. HAVE_AVX=
  247. HAVE_AVX2=
  248. HAVE_FMA3=
  249. include $(TOPDIR)/Makefile_kernel.conf
  250. endif
  251. endif
  252. ifndef NUM_PARALLEL
  253. NUM_PARALLEL = 1
  254. endif
  255. ifndef NUM_THREADS
  256. NUM_THREADS = $(NUM_CORES)
  257. endif
  258. ifeq ($(NUM_THREADS), 1)
  259. override USE_THREAD = 0
  260. override USE_OPENMP = 0
  261. endif
  262. ifdef USE_THREAD
  263. ifeq ($(USE_THREAD), 0)
  264. SMP =
  265. else
  266. SMP = 1
  267. endif
  268. else
  269. ifeq ($(NUM_THREADS), 1)
  270. SMP =
  271. else
  272. SMP = 1
  273. endif
  274. endif
  275. ifeq ($(SMP), 1)
  276. USE_LOCKING =
  277. endif
  278. ifndef NEED_PIC
  279. NEED_PIC = 1
  280. endif
  281. ARFLAGS =
  282. CPP = $(COMPILER) -E
  283. AR ?= $(CROSS_SUFFIX)ar
  284. AS ?= $(CROSS_SUFFIX)as
  285. LD ?= $(CROSS_SUFFIX)ld
  286. RANLIB ?= $(CROSS_SUFFIX)ranlib
  287. NM = $(CROSS_SUFFIX)nm
  288. DLLWRAP = $(CROSS_SUFFIX)dllwrap
  289. OBJCOPY = $(CROSS_SUFFIX)objcopy
  290. OBJCONV = $(CROSS_SUFFIX)objconv
  291. # When fortran support was either not detected or actively deselected, only build BLAS.
  292. ifeq ($(NOFORTRAN), 1)
  293. NO_LAPACK = 1
  294. override FEXTRALIB =
  295. endif
  296. ifeq ($(C_COMPILER), GCC)
  297. GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
  298. GCCVERSIONGT4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 4)
  299. GCCVERSIONGT5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 5)
  300. GCCVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 7)
  301. GCCVERSIONGTEQ8 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 8)
  302. GCCVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
  303. GCCVERSIONGTEQ11 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 11)
  304. GCCVERSIONGTEQ10 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 10)
  305. # Note that the behavior of -dumpversion is compile-time-configurable for
  306. # gcc-7.x and newer. Use -dumpfullversion there
  307. ifeq ($(GCCVERSIONGTEQ7),1)
  308. GCCDUMPVERSION_PARAM := -dumpfullversion
  309. else
  310. GCCDUMPVERSION_PARAM := -dumpversion
  311. endif
  312. GCCMINORVERSIONGTEQ1 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d.` \>= 1)
  313. GCCMINORVERSIONGTEQ2 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d.` \>= 2)
  314. GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d.` \>= 7)
  315. endif
  316. #
  317. # OS dependent settings
  318. #
  319. ifeq ($(OSNAME), Darwin)
  320. ifndef MACOSX_DEPLOYMENT_TARGET
  321. export MACOSX_DEPLOYMENT_TARGET=10.8
  322. endif
  323. MD5SUM = md5 -r
  324. endif
  325. ifneq (,$(findstring $(OSNAME), FreeBSD OpenBSD DragonFly))
  326. MD5SUM = md5 -r
  327. endif
  328. ifeq ($(OSNAME), NetBSD)
  329. MD5SUM = md5 -n
  330. endif
  331. ifeq ($(OSNAME), Linux)
  332. EXTRALIB += -lm
  333. NO_EXPRECISION = 1
  334. endif
  335. ifeq ($(OSNAME), Android)
  336. EXTRALIB += -lm
  337. endif
  338. ifeq ($(OSNAME), AIX)
  339. EXTRALIB += -lm
  340. endif
  341. ifeq ($(OSNAME), FreeBSD)
  342. ifeq ($(ARCH), $(filter $(ARCH),arm arm64))
  343. EXTRALIB += -lm
  344. endif
  345. endif
  346. ifeq ($(OSNAME), WINNT)
  347. NEED_PIC = 0
  348. NO_EXPRECISION = 1
  349. EXTRALIB += -defaultlib:advapi32
  350. SUFFIX = obj
  351. PSUFFIX = pobj
  352. LIBSUFFIX = a
  353. ifeq ($(C_COMPILER), CLANG)
  354. CCOMMON_OPT += -DMS_ABI
  355. endif
  356. #Version tests for supporting specific features (MS_ABI, POWER9 intrinsics)
  357. ifeq ($(GCCVERSIONGT4), 1)
  358. # GCC Major version > 4
  359. # It is compatible with MSVC ABI.
  360. CCOMMON_OPT += -DMS_ABI
  361. endif
  362. ifeq ($(GCCVERSIONGTEQ4), 1)
  363. ifeq ($(GCCMINORVERSIONGTEQ7), 1)
  364. # GCC Version >=4.7
  365. # It is compatible with MSVC ABI.
  366. CCOMMON_OPT += -DMS_ABI
  367. endif
  368. endif
  369. # Ensure the correct stack alignment on Win32
  370. # http://permalink.gmane.org/gmane.comp.lib.openblas.general/97
  371. ifeq ($(ARCH), x86)
  372. CCOMMON_OPT += -mincoming-stack-boundary=2
  373. FCOMMON_OPT += -mincoming-stack-boundary=2
  374. endif
  375. endif
  376. ifeq ($(OSNAME), Interix)
  377. NEED_PIC = 0
  378. NO_EXPRECISION = 1
  379. INTERIX_TOOL_DIR = /opt/gcc.3.3/i586-pc-interix3/bin
  380. endif
  381. ifeq ($(OSNAME), CYGWIN_NT)
  382. NEED_PIC = 0
  383. NO_EXPRECISION = 1
  384. OS_CYGWIN_NT = 1
  385. endif
  386. ifneq ($(OSNAME), WINNT)
  387. ifneq ($(OSNAME), CYGWIN_NT)
  388. ifneq ($(OSNAME), Interix)
  389. ifneq ($(OSNAME), Android)
  390. ifdef SMP
  391. EXTRALIB += -lpthread
  392. endif
  393. endif
  394. endif
  395. endif
  396. endif
  397. # ifeq logical or
  398. ifeq ($(OSNAME), $(filter $(OSNAME),WINNT CYGWIN_NT Interix))
  399. OS_WINDOWS=1
  400. endif
  401. ifdef QUAD_PRECISION
  402. CCOMMON_OPT += -DQUAD_PRECISION
  403. NO_EXPRECISION = 1
  404. endif
  405. ifneq ($(ARCH), x86)
  406. ifneq ($(ARCH), x86_64)
  407. NO_EXPRECISION = 1
  408. endif
  409. endif
  410. ifdef UTEST_CHECK
  411. CCOMMON_OPT += -DUTEST_CHECK
  412. SANITY_CHECK = 1
  413. endif
  414. ifdef SANITY_CHECK
  415. CCOMMON_OPT += -DSANITY_CHECK -DREFNAME=$(*F)f$(BU)
  416. endif
  417. MAX_STACK_ALLOC ?= 2048
  418. ifneq ($(MAX_STACK_ALLOC), 0)
  419. CCOMMON_OPT += -DMAX_STACK_ALLOC=$(MAX_STACK_ALLOC)
  420. endif
  421. ifdef USE_LOCKING
  422. ifneq ($(USE_LOCKING), 0)
  423. CCOMMON_OPT += -DUSE_LOCKING
  424. endif
  425. endif
  426. #
  427. # Architecture dependent settings
  428. #
  429. ifeq ($(ARCH), x86)
  430. ifndef BINARY
  431. NO_BINARY_MODE = 1
  432. endif
  433. ifeq ($(CORE), generic)
  434. NO_EXPRECISION = 1
  435. endif
  436. ifndef NO_EXPRECISION
  437. ifeq ($(F_COMPILER), GFORTRAN)
  438. # ifeq logical or. GCC or LSB
  439. ifeq ($(C_COMPILER), $(filter $(C_COMPILER),GCC LSB))
  440. EXPRECISION = 1
  441. CCOMMON_OPT += -DEXPRECISION -m128bit-long-double
  442. FCOMMON_OPT += -m128bit-long-double
  443. endif
  444. ifeq ($(C_COMPILER), CLANG)
  445. EXPRECISION = 1
  446. CCOMMON_OPT += -DEXPRECISION
  447. FCOMMON_OPT += -m128bit-long-double
  448. endif
  449. endif
  450. endif
  451. endif
  452. ifeq ($(ARCH), x86_64)
  453. ifeq ($(CORE), generic)
  454. NO_EXPRECISION = 1
  455. endif
  456. ifndef NO_EXPRECISION
  457. ifeq ($(F_COMPILER), GFORTRAN)
  458. # ifeq logical or. GCC or LSB
  459. ifeq ($(C_COMPILER), $(filter $(C_COMPILER),GCC LSB))
  460. EXPRECISION = 1
  461. CCOMMON_OPT += -DEXPRECISION -m128bit-long-double
  462. FCOMMON_OPT += -m128bit-long-double
  463. endif
  464. ifeq ($(C_COMPILER), CLANG)
  465. EXPRECISION = 1
  466. CCOMMON_OPT += -DEXPRECISION
  467. FCOMMON_OPT += -m128bit-long-double
  468. endif
  469. endif
  470. endif
  471. endif
  472. ifeq ($(C_COMPILER), INTEL)
  473. CCOMMON_OPT += -wd981
  474. endif
  475. ifeq ($(USE_OPENMP), 1)
  476. #check
  477. ifeq ($(USE_THREAD), 0)
  478. $(error OpenBLAS: Cannot set both USE_OPENMP=1 and USE_THREAD=0. The USE_THREAD=0 is only for building single thread version.)
  479. endif
  480. # ifeq logical or. GCC or LSB
  481. ifeq ($(C_COMPILER), $(filter $(C_COMPILER),GCC LSB))
  482. CCOMMON_OPT += -fopenmp
  483. endif
  484. ifeq ($(C_COMPILER), CLANG)
  485. CCOMMON_OPT += -fopenmp
  486. endif
  487. ifeq ($(C_COMPILER), INTEL)
  488. CCOMMON_OPT += -fopenmp
  489. endif
  490. ifeq ($(C_COMPILER), PGI)
  491. CCOMMON_OPT += -mp
  492. endif
  493. ifeq ($(C_COMPILER), OPEN64)
  494. CCOMMON_OPT += -mp
  495. CEXTRALIB += -lstdc++
  496. endif
  497. ifeq ($(C_COMPILER), PATHSCALE)
  498. CCOMMON_OPT += -mp
  499. endif
  500. endif
  501. ifeq ($(DYNAMIC_ARCH), 1)
  502. ifeq ($(ARCH), x86)
  503. DYNAMIC_CORE = KATMAI COPPERMINE NORTHWOOD PRESCOTT BANIAS \
  504. CORE2 PENRYN DUNNINGTON NEHALEM ATHLON OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
  505. endif
  506. ifeq ($(ARCH), x86_64)
  507. DYNAMIC_CORE = PRESCOTT CORE2
  508. ifeq ($(DYNAMIC_OLDER), 1)
  509. DYNAMIC_CORE += PENRYN DUNNINGTON
  510. endif
  511. DYNAMIC_CORE += NEHALEM
  512. ifeq ($(DYNAMIC_OLDER), 1)
  513. DYNAMIC_CORE += OPTERON OPTERON_SSE3
  514. endif
  515. DYNAMIC_CORE += BARCELONA
  516. ifeq ($(DYNAMIC_OLDER), 1)
  517. DYNAMIC_CORE += BOBCAT ATOM NANO
  518. endif
  519. ifneq ($(NO_AVX), 1)
  520. DYNAMIC_CORE += SANDYBRIDGE BULLDOZER PILEDRIVER STEAMROLLER EXCAVATOR
  521. endif
  522. ifneq ($(NO_AVX2), 1)
  523. DYNAMIC_CORE += HASWELL ZEN
  524. endif
  525. ifneq ($(NO_AVX512), 1)
  526. ifneq ($(NO_AVX2), 1)
  527. DYNAMIC_CORE += SKYLAKEX COOPERLAKE
  528. endif
  529. endif
  530. endif
  531. ifdef DYNAMIC_LIST
  532. override DYNAMIC_CORE = PRESCOTT $(DYNAMIC_LIST)
  533. XCCOMMON_OPT = -DDYNAMIC_LIST -DDYN_PRESCOTT
  534. XCCOMMON_OPT += $(foreach dcore,$(DYNAMIC_LIST),-DDYN_$(dcore))
  535. CCOMMON_OPT += $(XCCOMMON_OPT)
  536. #CCOMMON_OPT += -DDYNAMIC_LIST='$(DYNAMIC_LIST)'
  537. endif
  538. ifeq ($(ARCH), arm64)
  539. DYNAMIC_CORE = ARMV8
  540. DYNAMIC_CORE += CORTEXA53
  541. DYNAMIC_CORE += CORTEXA57
  542. DYNAMIC_CORE += CORTEXA72
  543. DYNAMIC_CORE += CORTEXA73
  544. DYNAMIC_CORE += NEOVERSEN1
  545. DYNAMIC_CORE += CORTEXA55
  546. DYNAMIC_CORE += FALKOR
  547. DYNAMIC_CORE += THUNDERX
  548. DYNAMIC_CORE += THUNDERX2T99
  549. DYNAMIC_CORE += TSV110
  550. DYNAMIC_CORE += EMAG8180
  551. DYNAMIC_CORE += THUNDERX3T110
  552. ifdef DYNAMIC_LIST
  553. override DYNAMIC_CORE = ARMV8 $(DYNAMIC_LIST)
  554. XCCOMMON_OPT = -DDYNAMIC_LIST -DDYN_ARMV8
  555. XCCOMMON_OPT += $(foreach dcore,$(DYNAMIC_LIST),-DDYN_$(dcore))
  556. endif
  557. endif
  558. ifeq ($(ARCH), mips64)
  559. DYNAMIC_CORE = LOONGSON3R3 LOONGSON3R4
  560. endif
  561. ifeq ($(ARCH), zarch)
  562. DYNAMIC_CORE = ZARCH_GENERIC
  563. # if the compiler accepts -march=arch11 or -march=z13 and can compile a file
  564. # with z13-specific inline assembly, then we can include support for Z13.
  565. # note: -march=z13 is equivalent to -march=arch11 yet some compiler releases
  566. # only support one or the other.
  567. # note: LLVM version 6.x supported -march=z13 yet could not handle vector
  568. # registers in inline assembly, so the check for supporting the -march flag is
  569. # not enough.
  570. ZARCH_TEST_COMPILE=-c $(TOPDIR)/kernel/zarch/damin_z13.c -I$(TOPDIR) -o /dev/null > /dev/null 2> /dev/null
  571. ZARCH_CC_SUPPORTS_ARCH11=$(shell $(CC) -march=arch11 $(ZARCH_TEST_COMPILE) && echo 1)
  572. ZARCH_CC_SUPPORTS_Z13=$(shell $(CC) -march=z13 $(ZARCH_TEST_COMPILE) && echo 1)
  573. ifeq ($(or $(ZARCH_CC_SUPPORTS_ARCH11), $(ZARCH_CC_SUPPORTS_Z13)), 1)
  574. DYNAMIC_CORE += Z13
  575. CCOMMON_OPT += -DDYN_Z13
  576. else
  577. $(info OpenBLAS: Not building Z13 kernels because the compiler $(CC) does not support it)
  578. endif
  579. # as above for z13, check for -march=arch12 and z14 support in the compiler.
  580. ZARCH_CC_SUPPORTS_ARCH12=$(shell $(CC) -march=arch12 $(ZARCH_TEST_COMPILE) && echo 1)
  581. ZARCH_CC_SUPPORTS_Z14=$(shell $(CC) -march=z14 $(ZARCH_TEST_COMPILE) && echo 1)
  582. ifeq ($(or $(ZARCH_CC_SUPPORTS_ARCH12), $(ZARCH_CC_SUPPORTS_Z14)), 1)
  583. DYNAMIC_CORE += Z14
  584. CCOMMON_OPT += -DDYN_Z14
  585. else
  586. $(info OpenBLAS: Not building Z14 kernels because the compiler $(CC) does not support it)
  587. endif
  588. endif # ARCH zarch
  589. ifeq ($(ARCH), power)
  590. ifneq ($(C_COMPILER), PGI)
  591. DYNAMIC_CORE = POWER6
  592. DYNAMIC_CORE += POWER8
  593. ifneq ($(C_COMPILER), GCC)
  594. DYNAMIC_CORE += POWER9
  595. DYNAMIC_CORE += POWER10
  596. CCOMMON_OPT += -DHAVE_P10_SUPPORT
  597. endif
  598. ifeq ($(C_COMPILER), GCC)
  599. ifeq ($(GCCVERSIONGT5), 1)
  600. DYNAMIC_CORE += POWER9
  601. else
  602. $(info, OpenBLAS: Your gcc version is too old to build the POWER9 kernels.)
  603. endif
  604. LDVERSIONGTEQ35 := $(shell expr `$(CC) -Wl,--version 2> /dev/null | head -1 | cut -f2 -d "." | cut -f1 -d "-"` \>= 35)
  605. ifeq ($(GCCVERSIONGTEQ11)$(LDVERSIONGTEQ35), 11)
  606. DYNAMIC_CORE += POWER10
  607. CCOMMON_OPT += -DHAVE_P10_SUPPORT
  608. else ifeq ($(GCCVERSIONGTEQ10), 1)
  609. ifeq ($(GCCMINORVERSIONGTEQ2)$(LDVERSIONGTEQ35), 11)
  610. DYNAMIC_CORE += POWER10
  611. CCOMMON_OPT += -DHAVE_P10_SUPPORT
  612. endif
  613. else
  614. $(info, OpenBLAS: Your gcc version is too old to build the POWER10 kernels.)
  615. endif
  616. endif
  617. else
  618. DYNAMIC_CORE = POWER8
  619. DYNAMIC_CORE += POWER9
  620. endif
  621. endif
  622. # If DYNAMIC_CORE is not set, DYNAMIC_ARCH cannot do anything, so force it to empty
  623. ifndef DYNAMIC_CORE
  624. override DYNAMIC_ARCH=
  625. endif
  626. endif
  627. ifeq ($(ARCH), ia64)
  628. NO_BINARY_MODE = 1
  629. BINARY_DEFINED = 1
  630. ifeq ($(F_COMPILER), GFORTRAN)
  631. ifeq ($(C_COMPILER), GCC)
  632. # EXPRECISION = 1
  633. # CCOMMON_OPT += -DEXPRECISION
  634. endif
  635. endif
  636. endif
  637. ifeq ($(ARCH), $(filter $(ARCH),mips64 mips))
  638. NO_BINARY_MODE = 1
  639. endif
  640. ifeq ($(ARCH), alpha)
  641. NO_BINARY_MODE = 1
  642. BINARY_DEFINED = 1
  643. endif
  644. ifeq ($(ARCH), arm)
  645. NO_BINARY_MODE = 1
  646. BINARY_DEFINED = 1
  647. CCOMMON_OPT += -marm
  648. FCOMMON_OPT += -marm
  649. # If softfp abi is mentioned on the command line, force it.
  650. ifeq ($(ARM_SOFTFP_ABI), 1)
  651. CCOMMON_OPT += -mfloat-abi=softfp
  652. FCOMMON_OPT += -mfloat-abi=softfp
  653. endif
  654. ifeq ($(OSNAME), Android)
  655. ifeq ($(ARM_SOFTFP_ABI), 1)
  656. EXTRALIB += -lm
  657. else
  658. EXTRALIB += -Wl,-lm_hard
  659. endif
  660. endif
  661. endif
  662. ifeq ($(ARCH), arm64)
  663. NO_BINARY_MODE = 1
  664. BINARY_DEFINED = 1
  665. ifdef INTERFACE64
  666. ifneq ($(INTERFACE64), 0)
  667. ifeq ($(F_COMPILER), GFORTRAN)
  668. FCOMMON_OPT += -fdefault-integer-8
  669. endif
  670. ifeq ($(F_COMPILER), FLANG)
  671. FCOMMON_OPT += -i8
  672. endif
  673. endif
  674. endif
  675. endif
  676. ifeq ($(ARCH), riscv64)
  677. NO_BINARY_MODE = 1
  678. BINARY_DEFINED = 1
  679. endif
  680. ifeq ($(ARCH), loongarch64)
  681. NO_BINARY_MODE = 1
  682. BINARY_DEFINED = 1
  683. endif
  684. #
  685. # C Compiler dependent settings
  686. #
  687. # ifeq logical or. GCC or CLANG or LSB
  688. # http://stackoverflow.com/questions/7656425/makefile-ifeq-logical-or
  689. ifeq ($(C_COMPILER), $(filter $(C_COMPILER),GCC CLANG LSB))
  690. CCOMMON_OPT += -Wall
  691. COMMON_PROF += -fno-inline
  692. NO_UNINITIALIZED_WARN = -Wno-uninitialized
  693. ifeq ($(QUIET_MAKE), 1)
  694. CCOMMON_OPT += $(NO_UNINITIALIZED_WARN) -Wno-unused
  695. endif
  696. ifdef NO_BINARY_MODE
  697. ifeq ($(ARCH), $(filter $(ARCH),mips64))
  698. ifdef BINARY64
  699. CCOMMON_OPT += -mabi=64
  700. else
  701. CCOMMON_OPT += -mabi=n32
  702. endif
  703. BINARY_DEFINED = 1
  704. else ifeq ($(ARCH), $(filter $(ARCH),mips))
  705. CCOMMON_OPT += -mabi=32
  706. BINARY_DEFINED = 1
  707. endif
  708. ifeq ($(CORE), $(filter $(CORE),LOONGSON3R3 LOONGSON3R4))
  709. CCOMMON_OPT += -march=loongson3a
  710. FCOMMON_OPT += -march=loongson3a
  711. endif
  712. ifeq ($(CORE), MIPS24K)
  713. CCOMMON_OPT += -mips32r2 -mtune=24kc $(MSA_FLAGS)
  714. FCOMMON_OPT += -mips32r2 -mtune=24kc $(MSA_FLAGS)
  715. endif
  716. ifeq ($(CORE), MIPS1004K)
  717. CCOMMON_OPT += -mips32r2 $(MSA_FLAGS)
  718. FCOMMON_OPT += -mips32r2 $(MSA_FLAGS)
  719. endif
  720. ifeq ($(CORE), P5600)
  721. CCOMMON_OPT += -mips32r5 -mnan=2008 -mtune=p5600 $(MSA_FLAGS)
  722. FCOMMON_OPT += -mips32r5 -mnan=2008 -mtune=p5600 $(MSA_FLAGS)
  723. endif
  724. ifeq ($(CORE), I6400)
  725. CCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=i6400 $(MSA_FLAGS)
  726. FCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=i6400 $(MSA_FLAGS)
  727. endif
  728. ifeq ($(CORE), P6600)
  729. CCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=p6600 $(MSA_FLAGS)
  730. FCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=p6600 $(MSA_FLAGS)
  731. endif
  732. ifeq ($(CORE), I6500)
  733. CCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=i6400 $(MSA_FLAGS)
  734. FCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=i6400 $(MSA_FLAGS)
  735. endif
  736. ifeq ($(OSNAME), AIX)
  737. BINARY_DEFINED = 1
  738. endif
  739. ifeq ($(ARCH), loongarch64)
  740. ifeq ($(CORE), LOONGSON3R5)
  741. CCOMMON_OPT += -march=loongarch64 -mabi=lp64
  742. FCOMMON_OPT += -march=loongarch64 -mabi=lp64
  743. endif
  744. endif
  745. endif
  746. ifndef BINARY_DEFINED
  747. ifneq ($(OSNAME), AIX)
  748. ifdef BINARY64
  749. ifneq ($(ARCH), riscv64)
  750. CCOMMON_OPT += -m64
  751. endif
  752. else
  753. CCOMMON_OPT += -m32
  754. endif
  755. endif
  756. endif
  757. endif
  758. ifeq ($(C_COMPILER), PGI)
  759. PGCVERSIONGT20 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -d "." -f 1` \> 20)
  760. PGCVERSIONGTEQ20 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -d "." -f 1` \>= 20)
  761. PGCMINORVERSIONGE11 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -c 4-5` == 11)
  762. PGCVERSIONCHECK := $(PGCVERSIONGT20)$(PGCVERSIONEQ20)$(PGCMINORVERSIONGE11)
  763. ifeq ($(PGCVERSIONCHECK), $(filter $(PGCVERSIONCHECK), 110 111 011))
  764. NEWPGI := 1
  765. endif
  766. ifdef BINARY64
  767. ifeq ($(ARCH), x86_64)
  768. CCOMMON_OPT += -tp p7-64
  769. ifneq ($(NEWPGI),1)
  770. CCOMMON_OPT += -D__MMX__ -Mnollvm
  771. endif
  772. else
  773. ifeq ($(ARCH), power)
  774. ifeq ($(CORE), POWER8)
  775. CCOMMON_OPT += -tp pwr8
  776. endif
  777. ifeq ($(CORE), POWER9)
  778. CCOMMON_OPT += -tp pwr9
  779. endif
  780. endif
  781. endif
  782. else
  783. CCOMMON_OPT += -tp p7
  784. endif
  785. endif
  786. ifeq ($(C_COMPILER), PATHSCALE)
  787. ifdef BINARY64
  788. CCOMMON_OPT += -m64
  789. else
  790. CCOMMON_OPT += -m32
  791. endif
  792. endif
  793. #
  794. # Fortran Compiler dependent settings
  795. #
  796. ifeq ($(F_COMPILER), NAG)
  797. FCOMMON_OPT += -dcfuns -recursive -ieee=full -w=obs -thread_safe
  798. ifdef INTERFACE64
  799. ifneq ($(INTERFACE64), 0)
  800. FCOMMON_OPT += -i8
  801. endif
  802. endif
  803. ifeq ($(USE_OPENMP), 1)
  804. FCOMMON_OPT += -openmp
  805. endif
  806. endif
  807. ifeq ($(F_COMPILER), FLANG)
  808. CCOMMON_OPT += -DF_INTERFACE_FLANG
  809. FCOMMON_OPT += -Mrecursive -Kieee
  810. ifeq ($(OSNAME), Linux)
  811. ifeq ($(ARCH), x86_64)
  812. FLANG_VENDOR := $(shell $(FC) --version|head -1 |cut -f 1 -d " ")
  813. ifeq ($(FLANG_VENDOR), AMD)
  814. FCOMMON_OPT += -fno-unroll-loops
  815. endif
  816. endif
  817. endif
  818. ifdef BINARY64
  819. ifdef INTERFACE64
  820. ifneq ($(INTERFACE64), 0)
  821. FCOMMON_OPT += -i8
  822. endif
  823. endif
  824. FCOMMON_OPT += -Wall
  825. else
  826. FCOMMON_OPT += -Wall
  827. endif
  828. ifeq ($(USE_OPENMP), 1)
  829. FCOMMON_OPT += -fopenmp
  830. endif
  831. endif
  832. ifeq ($(F_COMPILER), G77)
  833. CCOMMON_OPT += -DF_INTERFACE_G77
  834. FCOMMON_OPT += -Wall
  835. ifndef NO_BINARY_MODE
  836. ifneq ($(OSNAME), AIX)
  837. ifdef BINARY64
  838. FCOMMON_OPT += -m64
  839. else
  840. FCOMMON_OPT += -m32
  841. endif
  842. endif
  843. endif
  844. endif
  845. ifeq ($(F_COMPILER), G95)
  846. CCOMMON_OPT += -DF_INTERFACE_G95
  847. FCOMMON_OPT += -Wall
  848. ifneq ($(OSNAME), AIX)
  849. ifndef NO_BINARY_MODE
  850. ifdef BINARY64
  851. FCOMMON_OPT += -m64
  852. else
  853. FCOMMON_OPT += -m32
  854. endif
  855. endif
  856. ifneq ($(NO_LAPACKE), 1)
  857. FCOMMON_OPT += -fno-second-underscore
  858. endif
  859. endif
  860. endif
  861. ifeq ($(F_COMPILER), GFORTRAN)
  862. CCOMMON_OPT += -DF_INTERFACE_GFORT
  863. FCOMMON_OPT += -Wall
  864. # make single-threaded LAPACK calls thread-safe #1847
  865. FCOMMON_OPT += -frecursive
  866. # work around ABI problem with passing single-character arguments
  867. FCOMMON_OPT += -fno-optimize-sibling-calls
  868. #Don't include -lgfortran, when NO_LAPACK=1 or lsbcc
  869. ifneq ($(NO_LAPACK), 1)
  870. EXTRALIB += -lgfortran
  871. endif
  872. ifdef NO_BINARY_MODE
  873. ifeq ($(ARCH), $(filter $(ARCH),mips64))
  874. ifdef BINARY64
  875. FCOMMON_OPT += -mabi=64
  876. else
  877. FCOMMON_OPT += -mabi=n32
  878. endif
  879. else ifeq ($(ARCH), $(filter $(ARCH),mips))
  880. FCOMMON_OPT += -mabi=32
  881. endif
  882. else
  883. ifdef BINARY64
  884. ifneq ($(OSNAME), AIX)
  885. ifneq ($(ARCH), riscv64)
  886. FCOMMON_OPT += -m64
  887. endif
  888. endif
  889. ifdef INTERFACE64
  890. ifneq ($(INTERFACE64), 0)
  891. FCOMMON_OPT += -fdefault-integer-8
  892. endif
  893. endif
  894. else
  895. ifneq ($(OSNAME), AIX)
  896. FCOMMON_OPT += -m32
  897. endif
  898. endif
  899. endif
  900. ifeq ($(USE_OPENMP), 1)
  901. FCOMMON_OPT += -fopenmp
  902. endif
  903. endif
  904. ifeq ($(F_COMPILER), INTEL)
  905. CCOMMON_OPT += -DF_INTERFACE_INTEL
  906. ifdef INTERFACE64
  907. ifneq ($(INTERFACE64), 0)
  908. FCOMMON_OPT += -i8
  909. endif
  910. endif
  911. FCOMMON_OPT += -recursive -fp-model strict -assume protect-parens
  912. ifeq ($(USE_OPENMP), 1)
  913. FCOMMON_OPT += -fopenmp
  914. endif
  915. endif
  916. ifeq ($(F_COMPILER), FUJITSU)
  917. CCOMMON_OPT += -DF_INTERFACE_FUJITSU
  918. ifeq ($(USE_OPENMP), 1)
  919. FCOMMON_OPT += -openmp
  920. endif
  921. endif
  922. ifeq ($(F_COMPILER), IBM)
  923. CCOMMON_OPT += -DF_INTERFACE_IBM
  924. # FCOMMON_OPT += -qarch=440
  925. ifdef BINARY64
  926. FCOMMON_OPT += -q64
  927. ifdef INTERFACE64
  928. ifneq ($(INTERFACE64), 0)
  929. FCOMMON_OPT += -qintsize=8
  930. endif
  931. endif
  932. else
  933. FCOMMON_OPT += -q32
  934. endif
  935. ifeq ($(USE_OPENMP), 1)
  936. FCOMMON_OPT += -openmp
  937. endif
  938. endif
  939. ifeq ($(F_COMPILER), PGI)
  940. CCOMMON_OPT += -DF_INTERFACE_PGI
  941. COMMON_PROF += -DPGICOMPILER
  942. ifdef BINARY64
  943. ifdef INTERFACE64
  944. ifneq ($(INTERFACE64), 0)
  945. FCOMMON_OPT += -i8
  946. endif
  947. endif
  948. ifeq ($(ARCH), x86_64)
  949. FCOMMON_OPT += -tp p7-64
  950. else
  951. ifeq ($(ARCH), power)
  952. ifeq ($(CORE), POWER6)
  953. $(warning NVIDIA HPC compilers do not support POWER6.)
  954. endif
  955. ifeq ($(CORE), POWER8)
  956. FCOMMON_OPT += -tp pwr8
  957. endif
  958. ifeq ($(CORE), POWER9)
  959. FCOMMON_OPT += -tp pwr9
  960. endif
  961. ifeq ($(CORE), POWER10)
  962. $(warning NVIDIA HPC compilers do not support POWER10.)
  963. endif
  964. endif
  965. endif
  966. else
  967. FCOMMON_OPT += -tp p7
  968. endif
  969. FCOMMON_OPT += -Mrecursive -Kieee
  970. ifeq ($(USE_OPENMP), 1)
  971. FCOMMON_OPT += -mp
  972. endif
  973. endif
  974. ifeq ($(F_COMPILER), PATHSCALE)
  975. CCOMMON_OPT += -DF_INTERFACE_PATHSCALE
  976. ifdef BINARY64
  977. ifdef INTERFACE64
  978. ifneq ($(INTERFACE64), 0)
  979. FCOMMON_OPT += -i8
  980. endif
  981. endif
  982. endif
  983. ifeq ($(USE_OPENMP), 1)
  984. FCOMMON_OPT += -mp
  985. endif
  986. endif
  987. ifeq ($(F_COMPILER), OPEN64)
  988. CCOMMON_OPT += -DF_INTERFACE_OPEN64
  989. ifdef BINARY64
  990. ifdef INTERFACE64
  991. ifneq ($(INTERFACE64), 0)
  992. FCOMMON_OPT += -i8
  993. endif
  994. endif
  995. endif
  996. ifeq ($(ARCH), $(filter $(ARCH),mips64 mips))
  997. ifndef BINARY64
  998. FCOMMON_OPT += -n32
  999. else
  1000. FCOMMON_OPT += -n64
  1001. endif
  1002. ifeq ($(CORE), LOONGSON3R3)
  1003. FCOMMON_OPT += -loongson3 -static
  1004. endif
  1005. ifeq ($(CORE), LOONGSON3R4)
  1006. FCOMMON_OPT += -loongson3 -static
  1007. endif
  1008. else
  1009. ifndef BINARY64
  1010. FCOMMON_OPT += -m32
  1011. else
  1012. FCOMMON_OPT += -m64
  1013. endif
  1014. endif
  1015. ifeq ($(USE_OPENMP), 1)
  1016. FEXTRALIB += -lstdc++
  1017. FCOMMON_OPT += -mp
  1018. endif
  1019. endif
  1020. ifeq ($(C_COMPILER), OPEN64)
  1021. ifeq ($(ARCH), $(filter $(ARCH),mips64 mips))
  1022. ifndef BINARY64
  1023. CCOMMON_OPT += -n32
  1024. else
  1025. CCOMMON_OPT += -n64
  1026. endif
  1027. ifeq ($(CORE), LOONGSON3R3)
  1028. CCOMMON_OPT += -loongson3 -static
  1029. endif
  1030. ifeq ($(CORE), LOONGSON3R4)
  1031. CCOMMON_OPT += -loongson3 -static
  1032. endif
  1033. else
  1034. ifndef BINARY64
  1035. CCOMMON_OPT += -m32
  1036. else
  1037. CCOMMON_OPT += -m64
  1038. endif
  1039. endif
  1040. endif
  1041. ifeq ($(C_COMPILER), SUN)
  1042. CCOMMON_OPT += -w
  1043. ifeq ($(ARCH), x86)
  1044. CCOMMON_OPT += -m32
  1045. else
  1046. ifdef BINARY64
  1047. CCOMMON_OPT += -m64
  1048. else
  1049. CCOMMON_OPT += -m32
  1050. endif
  1051. endif
  1052. endif
  1053. ifeq ($(F_COMPILER), SUN)
  1054. CCOMMON_OPT += -DF_INTERFACE_SUN
  1055. FCOMMON_OPT += -ftrap=%none -xrecursive
  1056. ifeq ($(ARCH), x86)
  1057. FCOMMON_OPT += -m32
  1058. else
  1059. ifdef BINARY64
  1060. FCOMMON_OPT += -m64
  1061. else
  1062. FCOMMON_OPT += -m32
  1063. endif
  1064. endif
  1065. ifeq ($(USE_OPENMP), 1)
  1066. FCOMMON_OPT += -xopenmp=parallel
  1067. endif
  1068. endif
  1069. ifeq ($(F_COMPILER), COMPAQ)
  1070. CCOMMON_OPT += -DF_INTERFACE_COMPAQ
  1071. ifeq ($(USE_OPENMP), 1)
  1072. FCOMMON_OPT += -openmp
  1073. endif
  1074. endif
  1075. ifdef BINARY64
  1076. ifdef INTERFACE64
  1077. ifneq ($(INTERFACE64), 0)
  1078. CCOMMON_OPT +=
  1079. #-DUSE64BITINT
  1080. endif
  1081. endif
  1082. endif
  1083. ifeq ($(NEED_PIC), 1)
  1084. ifeq ($(C_COMPILER), IBM)
  1085. CCOMMON_OPT += -qpic=large
  1086. else
  1087. CCOMMON_OPT += -fPIC
  1088. endif
  1089. ifeq ($(F_COMPILER), SUN)
  1090. FCOMMON_OPT += -pic
  1091. else ifeq ($(F_COMPILER), NAG)
  1092. FCOMMON_OPT += -PIC
  1093. else
  1094. FCOMMON_OPT += -fPIC
  1095. endif
  1096. endif
  1097. ifeq ($(DYNAMIC_ARCH), 1)
  1098. CCOMMON_OPT += -DDYNAMIC_ARCH
  1099. endif
  1100. ifeq ($(DYNAMIC_OLDER), 1)
  1101. CCOMMON_OPT += -DDYNAMIC_OLDER
  1102. endif
  1103. ifeq ($(NO_LAPACK), 1)
  1104. CCOMMON_OPT += -DNO_LAPACK
  1105. #Disable LAPACK C interface
  1106. NO_LAPACKE = 1
  1107. endif
  1108. ifeq ($(NO_LAPACKE), 1)
  1109. CCOMMON_OPT += -DNO_LAPACKE
  1110. endif
  1111. ifeq ($(NO_AVX), 1)
  1112. CCOMMON_OPT += -DNO_AVX
  1113. endif
  1114. ifeq ($(ARCH), x86)
  1115. CCOMMON_OPT += -DNO_AVX
  1116. endif
  1117. ifeq ($(NO_AVX2), 1)
  1118. CCOMMON_OPT += -DNO_AVX2
  1119. endif
  1120. ifeq ($(NO_AVX512), 1)
  1121. CCOMMON_OPT += -DNO_AVX512
  1122. endif
  1123. ifdef SMP
  1124. CCOMMON_OPT += -DSMP_SERVER
  1125. ifeq ($(ARCH), mips64)
  1126. USE_SIMPLE_THREADED_LEVEL3 = 1
  1127. endif
  1128. ifeq ($(USE_OPENMP), 1)
  1129. # USE_SIMPLE_THREADED_LEVEL3 = 1
  1130. # NO_AFFINITY = 1
  1131. CCOMMON_OPT += -DUSE_OPENMP
  1132. endif
  1133. ifeq ($(BIGNUMA), 1)
  1134. CCOMMON_OPT += -DBIGNUMA
  1135. endif
  1136. endif
  1137. ifeq ($(NO_WARMUP), 1)
  1138. CCOMMON_OPT += -DNO_WARMUP
  1139. endif
  1140. ifeq ($(CONSISTENT_FPCSR), 1)
  1141. CCOMMON_OPT += -DCONSISTENT_FPCSR
  1142. endif
  1143. # Only for development
  1144. # CCOMMON_OPT += -DPARAMTEST
  1145. # CCOMMON_OPT += -DPREFETCHTEST
  1146. # CCOMMON_OPT += -DNO_SWITCHING
  1147. # USE_PAPI = 1
  1148. ifdef USE_PAPI
  1149. CCOMMON_OPT += -DUSE_PAPI
  1150. EXTRALIB += -lpapi -lperfctr
  1151. endif
  1152. ifdef BUFFERSIZE
  1153. CCOMMON_OPT += -DBUFFERSIZE=$(BUFFERSIZE)
  1154. endif
  1155. ifdef DYNAMIC_THREADS
  1156. CCOMMON_OPT += -DDYNAMIC_THREADS
  1157. endif
  1158. CCOMMON_OPT += -DMAX_CPU_NUMBER=$(NUM_THREADS)
  1159. CCOMMON_OPT += -DMAX_PARALLEL_NUMBER=$(NUM_PARALLEL)
  1160. ifdef USE_SIMPLE_THREADED_LEVEL3
  1161. CCOMMON_OPT += -DUSE_SIMPLE_THREADED_LEVEL3
  1162. endif
  1163. ifeq ($(USE_TLS), 1)
  1164. CCOMMON_OPT += -DUSE_TLS
  1165. endif
  1166. ifeq ($(BUILD_BFLOAT16), 1)
  1167. CCOMMON_OPT += -DBUILD_BFLOAT16
  1168. endif
  1169. ifeq ($(BUILD_SINGLE), 1)
  1170. CCOMMON_OPT += -DBUILD_SINGLE=1
  1171. endif
  1172. ifeq ($(BUILD_DOUBLE), 1)
  1173. CCOMMON_OPT += -DBUILD_DOUBLE=1
  1174. endif
  1175. ifeq ($(BUILD_COMPLEX), 1)
  1176. CCOMMON_OPT += -DBUILD_COMPLEX=1
  1177. endif
  1178. ifeq ($(BUILD_COMPLEX16), 1)
  1179. CCOMMON_OPT += -DBUILD_COMPLEX16=1
  1180. endif
  1181. CCOMMON_OPT += -DVERSION=\"$(VERSION)\"
  1182. ifndef SYMBOLPREFIX
  1183. SYMBOLPREFIX =
  1184. endif
  1185. ifndef SYMBOLSUFFIX
  1186. SYMBOLSUFFIX =
  1187. endif
  1188. ifndef LIBSONAMEBASE
  1189. LIBSONAMEBASE = openblas
  1190. endif
  1191. ifndef LIBNAMESUFFIX
  1192. LIBNAMEBASE = $(SYMBOLPREFIX)$(LIBSONAMEBASE)$(SYMBOLSUFFIX)
  1193. else
  1194. LIBNAMEBASE = $(SYMBOLPREFIX)$(LIBSONAMEBASE)$(SYMBOLSUFFIX)_$(LIBNAMESUFFIX)
  1195. endif
  1196. ifeq ($(OSNAME), CYGWIN_NT)
  1197. LIBPREFIX = cyg$(LIBNAMEBASE)
  1198. else
  1199. LIBPREFIX = lib$(LIBNAMEBASE)
  1200. endif
  1201. KERNELDIR = $(TOPDIR)/kernel/$(ARCH)
  1202. include $(TOPDIR)/Makefile.$(ARCH)
  1203. ifneq ($(C_COMPILER), PGI)
  1204. ifneq ($(C_COMPILER), SUN)
  1205. CCOMMON_OPT += -UASMNAME -UASMFNAME -UNAME -UCNAME -UCHAR_NAME -UCHAR_CNAME
  1206. endif
  1207. endif
  1208. CCOMMON_OPT += -DASMNAME=$(FU)$(*F) -DASMFNAME=$(FU)$(*F)$(BU) -DNAME=$(*F)$(BU) -DCNAME=$(*F) -DCHAR_NAME=\"$(*F)$(BU)\" -DCHAR_CNAME=\"$(*F)\"
  1209. ifeq ($(CORE), PPC440)
  1210. CCOMMON_OPT += -DALLOC_QALLOC
  1211. endif
  1212. ifeq ($(CORE), PPC440FP2)
  1213. STATIC_ALLOCATION = 1
  1214. endif
  1215. ifneq ($(OSNAME), Linux)
  1216. NO_AFFINITY = 1
  1217. endif
  1218. ifneq ($(ARCH), x86_64)
  1219. ifneq ($(ARCH), x86)
  1220. NO_AFFINITY = 1
  1221. endif
  1222. endif
  1223. ifdef NO_AFFINITY
  1224. ifeq ($(NO_AFFINITY), 0)
  1225. override undefine NO_AFFINITY
  1226. else
  1227. CCOMMON_OPT += -DNO_AFFINITY
  1228. endif
  1229. endif
  1230. ifdef FUNCTION_PROFILE
  1231. CCOMMON_OPT += -DFUNCTION_PROFILE
  1232. endif
  1233. ifdef HUGETLB_ALLOCATION
  1234. CCOMMON_OPT += -DALLOC_HUGETLB
  1235. endif
  1236. ifdef HUGETLBFILE_ALLOCATION
  1237. CCOMMON_OPT += -DALLOC_HUGETLBFILE -DHUGETLB_FILE_NAME=$(HUGETLBFILE_ALLOCATION)
  1238. endif
  1239. ifdef STATIC_ALLOCATION
  1240. CCOMMON_OPT += -DALLOC_STATIC
  1241. endif
  1242. ifdef DEVICEDRIVER_ALLOCATION
  1243. CCOMMON_OPT += -DALLOC_DEVICEDRIVER -DDEVICEDRIVER_NAME=\"/dev/mapper\"
  1244. endif
  1245. ifdef MIXED_MEMORY_ALLOCATION
  1246. CCOMMON_OPT += -DMIXED_MEMORY_ALLOCATION
  1247. endif
  1248. ifeq ($(OSNAME), SunOS)
  1249. TAR = gtar
  1250. PATCH = gpatch
  1251. GREP = ggrep
  1252. AWK = nawk
  1253. else
  1254. TAR = tar
  1255. PATCH = patch
  1256. GREP = grep
  1257. AWK = awk
  1258. endif
  1259. ifndef MD5SUM
  1260. MD5SUM = md5sum
  1261. endif
  1262. REVISION = -r$(VERSION)
  1263. MAJOR_VERSION = $(word 1,$(subst ., ,$(VERSION)))
  1264. ifeq ($(DEBUG), 1)
  1265. COMMON_OPT += -g
  1266. endif
  1267. ifeq ($(DEBUG), 1)
  1268. FCOMMON_OPT += -g
  1269. endif
  1270. ifndef COMMON_OPT
  1271. COMMON_OPT = -O2
  1272. endif
  1273. ifndef FCOMMON_OPT
  1274. FCOMMON_OPT = -O2 -frecursive
  1275. endif
  1276. override CFLAGS += $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR)
  1277. override PFLAGS += $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR) -DPROFILE $(COMMON_PROF)
  1278. override FFLAGS += $(COMMON_OPT) $(FCOMMON_OPT)
  1279. override FPFLAGS += $(FCOMMON_OPT) $(COMMON_PROF)
  1280. #MAKEOVERRIDES =
  1281. ifdef NEED_PIC
  1282. ifeq (,$(findstring PIC,$(FFLAGS)))
  1283. override FFLAGS += -fPIC
  1284. endif
  1285. endif
  1286. #For LAPACK Fortran codes.
  1287. #Disable -fopenmp for LAPACK Fortran codes on Windows.
  1288. ifdef OS_WINDOWS
  1289. LAPACK_FFLAGS := $(filter-out -fopenmp -mp -openmp -xopenmp=parallel,$(FFLAGS))
  1290. LAPACK_FPFLAGS := $(filter-out -fopenmp -mp -openmp -xopenmp=parallel,$(FPFLAGS))
  1291. else
  1292. LAPACK_FFLAGS := $(FFLAGS)
  1293. LAPACK_FPFLAGS := $(FPFLAGS)
  1294. endif
  1295. ifeq ($(F_COMPILER),NAG)
  1296. LAPACK_FFLAGS := $(filter-out -msse3 -mssse3 -msse4.1 -mavx -mavx2 -mskylake-avx512 ,$(FFLAGS))
  1297. endif
  1298. LAPACK_CFLAGS = $(CFLAGS)
  1299. LAPACK_CFLAGS += -DHAVE_LAPACK_CONFIG_H
  1300. ifdef INTERFACE64
  1301. ifneq ($(INTERFACE64), 0)
  1302. LAPACK_CFLAGS += -DLAPACK_ILP64
  1303. endif
  1304. endif
  1305. ifdef OS_WINDOWS
  1306. LAPACK_CFLAGS += -DOPENBLAS_OS_WINDOWS
  1307. endif
  1308. ifeq ($(C_COMPILER), LSB)
  1309. LAPACK_CFLAGS += -DLAPACK_COMPLEX_STRUCTURE
  1310. endif
  1311. ifndef SUFFIX
  1312. SUFFIX = o
  1313. endif
  1314. ifndef PSUFFIX
  1315. PSUFFIX = po
  1316. endif
  1317. ifndef LIBSUFFIX
  1318. LIBSUFFIX = a
  1319. endif
  1320. ifneq ($(DYNAMIC_ARCH), 1)
  1321. ifndef SMP
  1322. LIBNAME = $(LIBPREFIX)_$(LIBCORE)$(REVISION).$(LIBSUFFIX)
  1323. LIBNAME_P = $(LIBPREFIX)_$(LIBCORE)$(REVISION)_p.$(LIBSUFFIX)
  1324. else
  1325. LIBNAME = $(LIBPREFIX)_$(LIBCORE)p$(REVISION).$(LIBSUFFIX)
  1326. LIBNAME_P = $(LIBPREFIX)_$(LIBCORE)p$(REVISION)_p.$(LIBSUFFIX)
  1327. endif
  1328. else
  1329. ifndef SMP
  1330. LIBNAME = $(LIBPREFIX)$(REVISION).$(LIBSUFFIX)
  1331. LIBNAME_P = $(LIBPREFIX)$(REVISION)_p.$(LIBSUFFIX)
  1332. else
  1333. LIBNAME = $(LIBPREFIX)p$(REVISION).$(LIBSUFFIX)
  1334. LIBNAME_P = $(LIBPREFIX)p$(REVISION)_p.$(LIBSUFFIX)
  1335. endif
  1336. endif
  1337. LIBDLLNAME = $(LIBPREFIX).dll
  1338. IMPLIBNAME = lib$(LIBNAMEBASE).dll.a
  1339. ifneq ($(OSNAME), AIX)
  1340. LIBSONAME = $(LIBNAME:.$(LIBSUFFIX)=.so)
  1341. else
  1342. LIBSONAME = $(LIBNAME:.$(LIBSUFFIX)=.a)
  1343. endif
  1344. LIBDYNNAME = $(LIBNAME:.$(LIBSUFFIX)=.dylib)
  1345. LIBDEFNAME = $(LIBNAME:.$(LIBSUFFIX)=.def)
  1346. LIBEXPNAME = $(LIBNAME:.$(LIBSUFFIX)=.exp)
  1347. LIBZIPNAME = $(LIBNAME:.$(LIBSUFFIX)=.zip)
  1348. LIBS = $(TOPDIR)/$(LIBNAME)
  1349. LIBS_P = $(TOPDIR)/$(LIBNAME_P)
  1350. LIB_COMPONENTS = BLAS
  1351. ifneq ($(NO_CBLAS), 1)
  1352. LIB_COMPONENTS += CBLAS
  1353. endif
  1354. ifneq ($(NO_LAPACK), 1)
  1355. LIB_COMPONENTS += LAPACK
  1356. ifneq ($(NO_LAPACKE), 1)
  1357. LIB_COMPONENTS += LAPACKE
  1358. endif
  1359. ifeq ($(BUILD_RELAPACK), 1)
  1360. LIB_COMPONENTS += ReLAPACK
  1361. endif
  1362. endif
  1363. ifeq ($(ONLY_CBLAS), 1)
  1364. LIB_COMPONENTS = CBLAS
  1365. endif
  1366. export OSNAME
  1367. export ARCH
  1368. export CORE
  1369. export LIBCORE
  1370. export __BYTE_ORDER__
  1371. export ELF_VERSION
  1372. export PGCPATH
  1373. export CONFIG
  1374. export CC
  1375. export FC
  1376. export BU
  1377. export FU
  1378. export NEED2UNDERSCORES
  1379. export USE_THREAD
  1380. export NUM_THREADS
  1381. export NUM_CORES
  1382. export SMP
  1383. export MAKEFILE_RULE
  1384. export NEED_PIC
  1385. export BINARY
  1386. export BINARY32
  1387. export BINARY64
  1388. export F_COMPILER
  1389. export C_COMPILER
  1390. export USE_OPENMP
  1391. export CROSS
  1392. export CROSS_SUFFIX
  1393. export NOFORTRAN
  1394. export NO_FBLAS
  1395. export EXTRALIB
  1396. export CEXTRALIB
  1397. export FEXTRALIB
  1398. export HAVE_SSE
  1399. export HAVE_SSE2
  1400. export HAVE_SSE3
  1401. export HAVE_SSSE3
  1402. export HAVE_SSE4_1
  1403. export HAVE_SSE4_2
  1404. export HAVE_SSE4A
  1405. export HAVE_SSE5
  1406. export HAVE_AVX
  1407. export HAVE_AVX2
  1408. export HAVE_FMA3
  1409. export HAVE_VFP
  1410. export HAVE_VFPV3
  1411. export HAVE_VFPV4
  1412. export HAVE_NEON
  1413. export HAVE_MSA
  1414. export MSA_FLAGS
  1415. export KERNELDIR
  1416. export FUNCTION_PROFILE
  1417. export TARGET_CORE
  1418. export NO_AVX512
  1419. export NO_AVX2
  1420. export BUILD_BFLOAT16
  1421. export SBGEMM_UNROLL_M
  1422. export SBGEMM_UNROLL_N
  1423. export SGEMM_UNROLL_M
  1424. export SGEMM_UNROLL_N
  1425. export DGEMM_UNROLL_M
  1426. export DGEMM_UNROLL_N
  1427. export QGEMM_UNROLL_M
  1428. export QGEMM_UNROLL_N
  1429. export CGEMM_UNROLL_M
  1430. export CGEMM_UNROLL_N
  1431. export ZGEMM_UNROLL_M
  1432. export ZGEMM_UNROLL_N
  1433. export XGEMM_UNROLL_M
  1434. export XGEMM_UNROLL_N
  1435. export CGEMM3M_UNROLL_M
  1436. export CGEMM3M_UNROLL_N
  1437. export ZGEMM3M_UNROLL_M
  1438. export ZGEMM3M_UNROLL_N
  1439. export XGEMM3M_UNROLL_M
  1440. export XGEMM3M_UNROLL_N
  1441. ifdef USE_CUDA
  1442. export CUDADIR
  1443. export CUCC
  1444. export CUFLAGS
  1445. export CULIB
  1446. endif
  1447. .SUFFIXES: .$(PSUFFIX) .$(SUFFIX) .f
  1448. .f.$(SUFFIX):
  1449. $(FC) $(FFLAGS) -c $< -o $(@F)
  1450. .f.$(PSUFFIX):
  1451. $(FC) $(FPFLAGS) -pg -c $< -o $(@F)
  1452. ifdef BINARY64
  1453. PATHSCALEPATH = /opt/pathscale/lib/3.1
  1454. PGIPATH = /opt/pgi/linux86-64/7.1-5/lib
  1455. else
  1456. PATHSCALEPATH = /opt/pathscale/lib/3.1/32
  1457. PGIPATH = /opt/pgi/linux86/7.1-5/lib
  1458. endif
  1459. ACMLPATH = /opt/acml/4.3.0
  1460. ifneq ($(OSNAME), Darwin)
  1461. MKLPATH = /opt/intel/mkl/10.2.2.025/lib
  1462. else
  1463. MKLPATH = /Library/Frameworks/Intel_MKL.framework/Versions/10.0.1.014/lib
  1464. endif
  1465. ATLASPATH = /opt/atlas/3.9.17/opteron
  1466. FLAMEPATH = $(HOME)/flame/lib
  1467. ifneq ($(OSNAME), SunOS)
  1468. SUNPATH = /opt/sunstudio12.1
  1469. else
  1470. SUNPATH = /opt/SUNWspro
  1471. endif