Merge pull request #110 from xianyi/develop

rebase

Author: martin-frbg

Makefile.system

@@ -93,6 +93,12 @@ endif
 ifdef TARGET
 GETARCH_FLAGS := -DFORCE_$(TARGET)
 GETARCH_FLAGS += -DUSER_TARGET
+ifeq ($(TARGET), GENERIC)
+ifeq ($(DYNAMIC_ARCH), 1)
+override NO_EXPRECISION=1
+export NO_EXPRECiSION
+endif
+endif
 endif
 
 # Force fallbacks for 32bit

cmake/os.cmake

@@ -84,6 +84,14 @@ if (X86)
   set(NO_EXPRECISION 1)
 endif ()
 
+if (DYNAMIC_ARCH)
+if (TARGET)
+if (${TARGET} STREQUAL "GENERIC")
+  set(NO_EXPRECISION 1)
+endif ()
+endif ()
+endif ()
+
 if (UTEST_CHECK)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DUTEST_CHECK")
   set(SANITY_CHECK 1)

cmake/prebuild.cmake

@@ -139,6 +139,36 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSS
       set(CGEMM3M_UNROLL_N 4)
       set(ZGEMM3M_UNROLL_M 4)
       set(ZGEMM3M_UNROLL_N 4)
+  elseif ("${TCORE}" STREQUAL "BARCELONA")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define HAVE_SSE3\n")
+  elseif ("${TCORE}" STREQUAL "STEAMROLLER")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define HAVE_SSE3\n")
+  elseif ("${TCORE}" STREQUAL "EXCAVATOR")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define HAVE_SSE3\n")
+  elseif ("${TCORE}" STREQUAL "NEHALEM")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define HAVE_SSE3\n")
+  elseif ("${TCORE}" STREQUAL "PRESCOTT")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define HAVE_SSE3\n")
+  elseif ("${TCORE}" STREQUAL "SANDYBRIDGE")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define HAVE_AVX\n")
+  elseif ("${TCORE}" STREQUAL "HASWELL")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define HAVE_AVX2\n")
+  elseif ("${TCORE}" STREQUAL "ZEN")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define HAVE_AVX2\n")
+  elseif ("${TCORE}" STREQUAL "SKYLAKEX")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define HAVE_AVX512\n")
+  elseif ("${TCORE}" STREQUAL "COOPERLAKE")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define HAVE_AVX512\n")
   elseif ("${TCORE}" STREQUAL "ARMV7")
     file(APPEND ${TARGET_CONF_TEMP}
       "#define L1_DATA_SIZE\t65536\n"

cmake/system.cmake

@@ -64,12 +64,36 @@ if (DEFINED TARGET)
     if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU")
       execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
       if (${GCC_VERSION} VERSION_GREATER 4.7 OR ${GCC_VERSION} VERSION_EQUAL 4.7)
-        set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mavx2")
+        set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3 -mavx2")
       endif()
     elseif (${CMAKE_C_COMPILER_ID} STREQUAL "CLANG")
-      set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mavx2")
+      set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse -msse3 -mavx2")
     endif()
   endif()
+  if (${TARGET} STREQUAL "HASWELL" AND NOT NO_AVX2)
+      set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3 -mavx2")
+  endif()
+  if (${TARGET} STREQUAL "ZEN" AND NOT NO_AVX2)
+      set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3 -mavx2")
+  endif()
+  if (${TARGET} STREQUAL "SANDYBRIDGE" AND NOT NO_AVX)
+      set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3 -mavx")
+  endif()
+  if (${TARGET} STREQUAL "BARCELONA" OR ${TARGET} STREQUAL "STEAMROLLER" OR ${TARGET} STREQUAL "BULLDOZER" OR ${TARGET} STREQUAL "EXCAVATOR")
+      set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3")
+  endif()
+  if (${TARGET} STREQUAL "PILEDRIVER" OR ${TARGET} STREQUAL "BOBCAT" OR ${TARGET} STREQUAL "OPTERON_SSE3")
+      set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3")
+  endif()
+  if (${TARGET} STREQUAL "PRESCOTT" OR ${TARGET} STREQUAL "NANO")
+      set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3")
+  endif()
+  if (${TARGET} STREQUAL "NEHALEM" OR ${TARGET} STREQUAL "ATOM")
+      set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3")
+  endif()
+  if (${TARGET} STREQUAL "CORE2" OR ${TARGET} STREQUAL "PENRYN" OR ${TARGET} STREQUAL "DUNNINGTON")
+      set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3")
+  endif()
   if (DEFINED HAVE_SSE)
     set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse")
   endif()

common_arm64.h

@@ -142,14 +142,8 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #define HUGE_PAGESIZE   ( 4 << 20)
 
 #ifndef BUFFERSIZE
-#if defined(CORTEXA57)
-#define BUFFER_SIZE     (20 << 20)
-#elif defined(TSV110) || defined(EMAG8180)
 #define BUFFER_SIZE     (32 << 20)
 #else
-#define BUFFER_SIZE     (16 << 20)
-#endif
-#else
 #define BUFFER_SIZE	(32 << BUFFERSIZE)
 #endif
 

f_check

@@ -33,7 +33,7 @@ if ($compiler eq "") {
               "ppuf77", "ppuf95", "ppuf90", "ppuxlf",
 	      "pathf90", "pathf95",
 	      "pgf95", "pgf90", "pgf77",
-	      "flang",
+	      "flang", "egfortran",
               "ifort");
 
 OUTER:

kernel/arm/zdot.c

@@ -73,7 +73,7 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
 		i++ ;
 
 	}
-#if !defined(__POWER__)	
+#if !defined(__PPC__)	
         CREAL(result) = dot[0];
 	CIMAG(result) = dot[1];
 #else

kernel/power/KERNEL.POWER10

@@ -34,12 +34,12 @@ SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
 SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 DGEMMKERNEL    =  dgemm_kernel_power10.c
-DGEMMINCOPY    = ../generic/gemm_ncopy_16.c
-DGEMMITCOPY    =  dgemm_tcopy_16_power8.S
-DGEMMONCOPY    =  dgemm_ncopy_4_power8.S
-DGEMMOTCOPY    =  ../generic/gemm_tcopy_4.c
-DGEMMINCOPYOBJ =  dgemm_incopy$(TSUFFIX).$(SUFFIX)
-DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
+DGEMMINCOPY    =
+DGEMMITCOPY    =
+DGEMMONCOPY    =  dgemm_ncopy_8_power10.c
+DGEMMOTCOPY    =  ../generic/gemm_tcopy_8.c
+DGEMMINCOPYOBJ =
+DGEMMITCOPYOBJ =
 DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
 DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
@@ -69,7 +69,7 @@ STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
 STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
 
 DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
-DTRSMKERNEL_LT	= dtrsm_kernel_LT_16x4_power8.S
+DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
 DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
 DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c