Merge pull request #941 from sva-img/develop

MIPS n32 ABI support, MSA support detection and rename ARCH, ARCHFLAGS

Makefile.prebuild

@@ -17,14 +17,26 @@ ifdef CPUIDEMU
 EXFLAGS = -DCPUIDEMU -DVENDOR=99
 endif
 
+ifeq ($(TARGET), P5600)
+TARGET_FLAGS = -mips32r5
+endif
+
+ifeq ($(TARGET), I6400)
+TARGET_FLAGS = -mips64r6
+endif
+
+ifeq ($(TARGET), P6600)
+TARGET_FLAGS = -mips64r6
+endif
+
 all: getarch_2nd
 	./getarch_2nd  0 >> $(TARGET_MAKE)
 	./getarch_2nd  1 >> $(TARGET_CONF)
 
 config.h : c_check f_check getarch
-	perl ./c_check $(TARGET_MAKE) $(TARGET_CONF) $(CC)
+	perl ./c_check $(TARGET_MAKE) $(TARGET_CONF) $(CC) $(TARGET_FLAGS)
 ifneq ($(ONLY_CBLAS), 1)
-	perl ./f_check $(TARGET_MAKE) $(TARGET_CONF) $(FC)
+	perl ./f_check $(TARGET_MAKE) $(TARGET_CONF) $(FC) $(TARGET_FLAGS)
 else
 #When we only build CBLAS, we set NOFORTRAN=2
 	echo "NOFORTRAN=2" >> $(TARGET_MAKE)

Makefile.system

@@ -159,7 +159,7 @@ ifndef GOTOBLAS_MAKEFILE
 export GOTOBLAS_MAKEFILE = 1
 
 # Generating Makefile.conf and config.h
-DUMMY := $(shell $(MAKE) -C $(TOPDIR) -f Makefile.prebuild CC="$(CC)" FC="$(FC)" HOSTCC="$(HOSTCC)" CFLAGS="$(GETARCH_FLAGS)" BINARY=$(BINARY) USE_OPENMP=$(USE_OPENMP) TARGET_CORE=$(TARGET_CORE) ONLY_CBLAS=$(ONLY_CBLAS) all)
+DUMMY := $(shell $(MAKE) -C $(TOPDIR) -f Makefile.prebuild CC="$(CC)" FC="$(FC)" HOSTCC="$(HOSTCC)" CFLAGS="$(GETARCH_FLAGS)" BINARY=$(BINARY) USE_OPENMP=$(USE_OPENMP) TARGET_CORE=$(TARGET_CORE) ONLY_CBLAS=$(ONLY_CBLAS) TARGET=$(TARGET) all)
 
 ifndef TARGET_CORE
 include $(TOPDIR)/Makefile.conf
@@ -502,13 +502,16 @@ endif
 
 ifdef NO_BINARY_MODE
 
-ifeq ($(ARCH), $(filter $(ARCH),mips64 mips))
+ifeq ($(ARCH), $(filter $(ARCH),mips64))
 ifdef BINARY64
 CCOMMON_OPT += -mabi=64
 else
-CCOMMON_OPT += -mabi=32
+CCOMMON_OPT += -mabi=n32
 endif
 BINARY_DEFINED = 1
+else ifeq ($(ARCH), $(filter $(ARCH),mips))
+CCOMMON_OPT += -mabi=32
+BINARY_DEFINED = 1
 endif
 
 ifeq ($(CORE), LOONGSON3A)
@@ -522,13 +525,18 @@ FCOMMON_OPT += -march=mips64
 endif
 
 ifeq ($(CORE), P5600)
-CCOMMON_OPT += -mmsa
-FCOMMON_OPT += -mmsa
+CCOMMON_OPT += -mips32r5 -mnan=2008 -mtune=p5600 $(MSA_FLAGS)
+FCOMMON_OPT += -mips32r5 -mnan=2008 -mtune=p5600 $(MSA_FLAGS)
+endif
+
+ifeq ($(CORE), I6400)
+CCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=i6400 $(MSA_FLAGS)
+FCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=i6400 $(MSA_FLAGS)
 endif
 
-ifneq ($(filter $(CORE), I6400 P6600),)
-CCOMMON_OPT += -mmsa
-FCOMMON_OPT += -mmsa
+ifeq ($(CORE), P6600)
+CCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=p6600 $(MSA_FLAGS)
+FCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=p6600 $(MSA_FLAGS)
 endif
 
 ifeq ($(OSNAME), AIX)
@@ -599,12 +607,14 @@ ifneq ($(NO_LAPACK), 1)
 EXTRALIB += -lgfortran
 endif
 ifdef NO_BINARY_MODE
-ifeq ($(ARCH), $(filter $(ARCH),mips64 mips))
+ifeq ($(ARCH), $(filter $(ARCH),mips64))
 ifdef BINARY64
 FCOMMON_OPT += -mabi=64
 else
-FCOMMON_OPT += -mabi=32
+FCOMMON_OPT += -mabi=n32
 endif
+else ifeq ($(ARCH), $(filter $(ARCH),mips))
+FCOMMON_OPT += -mabi=32
 endif
 else
 ifdef BINARY64
@@ -687,20 +697,6 @@ FCOMMON_OPT += -i8
 endif
 endif
 endif
-
-ifeq ($(filter $(ARCH),mips64 mips))
-ifndef BINARY64
-FCOMMON_OPT += -m32
-else
-FCOMMON_OPT += -m64
-endif
-else
-ifdef BINARY64
-FCOMMON_OPT += -mabi=64
-else
-FCOMMON_OPT += -mabi=32
-endif
-endif
  
 ifeq ($(USE_OPENMP), 1)
 FCOMMON_OPT += -mp
@@ -1136,6 +1132,8 @@ export HAVE_VFP
 export HAVE_VFPV3
 export HAVE_VFPV4
 export HAVE_NEON
+export HAVE_MSA
+export MSA_FLAGS
 export KERNELDIR
 export FUNCTION_PROFILE
 export TARGET_CORE

c_check

@@ -1,6 +1,7 @@
 #!/usr/bin/perl
 
 use File::Basename;
+use File::Temp qw(tempfile);
 
 # Checking cross compile
 $hostos   = `uname -s | sed -e s/\-.*//`;    chop($hostos);
@@ -10,6 +11,7 @@ $hostarch = "arm" if ($hostarch =~ /^arm.*/);
 $hostarch = "arm64" if ($hostarch eq "aarch64");
 $hostarch = "power" if ($hostarch =~ /^(powerpc|ppc).*/);
 
+$tmpf = new File::Temp( UNLINK => 1 );
 $binary = $ENV{"BINARY"};
 
 $makefile = shift(@ARGV);
@@ -79,10 +81,15 @@ if ($os eq "AIX") {
     $defined = 1;
 }
 
-if (($architecture eq "mips") || ($architecture eq "mips64")) {
-     $compiler_name .= " -mabi=32" if ($binary eq "32");   
-     $compiler_name .= " -mabi=64" if ($binary eq "64");
-     $defined = 1;
+if ($architecture eq "mips") {
+    $compiler_name .= " -mabi=32";
+    $defined = 1;
+}
+
+if ($architecture eq "mips64") {
+    $compiler_name .= " -mabi=n32" if ($binary eq "32");
+    $compiler_name .= " -mabi=64" if ($binary eq "64");
+    $defined = 1;
 }
 
 if (($architecture eq "arm") || ($architecture eq "arm64")) {
@@ -152,6 +159,24 @@ if ($?) {
     die 1;
 }
 
+$have_msa = 0;
+if (($architecture eq "mips") || ($architecture eq "mips64")) {
+    $code = '"addvi.b $w0, $w1, 1"';
+    $msa_flags = "-mmsa -mfp64 -msched-weight -mload-store-pairs";
+    print $tmpf "#include <msa.h>\n\n";
+    print $tmpf "void main(void){ __asm__ volatile($code); }\n";
+
+    $args = "$msa_flags -o $tmpf.o -x c $tmpf";
+    my @cmd = ("$compiler_name $args");
+    system(@cmd) == 0;
+    if ($? != 0) {
+        $have_msa = 0;
+    } else {
+        $have_msa = 1;
+    }
+    unlink("$tmpf.o");
+}
+
 $architecture = x86    if ($data =~ /ARCH_X86/);
 $architecture = x86_64 if ($data =~ /ARCH_X86_64/);
 $architecture = power  if ($data =~ /ARCH_POWER/);
@@ -246,6 +271,8 @@ print MAKEFILE "FU=$need_fu\n" if $need_fu ne "";
 print MAKEFILE "CROSS_SUFFIX=$cross_suffix\n" if $cross != 0 && $cross_suffix ne "";
 print MAKEFILE "CROSS=1\n" if $cross != 0;
 print MAKEFILE "CEXTRALIB=$linker_L $linker_l $linker_a\n";
+print MAKEFILE "HAVE_MSA=1\n" if $have_msa eq 1;
+print MAKEFILE "MSA_FLAGS=$msa_flags\n" if $have_msa eq 1;
 
 $os           =~ tr/[a-z]/[A-Z]/;
 $architecture =~ tr/[a-z]/[A-Z]/;
@@ -257,6 +284,7 @@ print CONFFILE "#define C_$compiler\t1\n";
 print CONFFILE "#define __32BIT__\t1\n"  if $binformat eq bin32;
 print CONFFILE "#define __64BIT__\t1\n"  if $binformat eq bin64;
 print CONFFILE "#define FUNDERSCORE\t$need_fu\n" if $need_fu ne "";
+print CONFFILE "#define HAVE_MSA\t1\n"  if $have_msa eq 1;
 
 if ($os eq "LINUX") {
 

f_check

@@ -223,7 +223,12 @@ if (!$?) {
 	}
        #For gfortran MIPS
 	if ($?) {
-      $link = `$compiler $openmp -mabi=32 -v ftest2.f 2>&1 && rm -f a.out a.exe`;
+    $mips_data = `$compiler_bin -E -dM - < /dev/null`;
+    if ($mips_data =~ /_MIPS_ISA_MIPS64/) {
+        $link = `$compiler $openmp -mabi=n32 -v ftest2.f 2>&1 && rm -f a.out a.exe`;
+    } else {
+        $link = `$compiler $openmp -mabi=32 -v ftest2.f 2>&1 && rm -f a.out a.exe`;
+    }
 	}
 	$binary = "" if ($?);
     }

kernel/mips/KERNEL.P5600

@@ -30,10 +30,17 @@ IDMAXKERNEL  = ../mips/imax.c
 ISMINKERNEL  = ../mips/imin.c
 IDMINKERNEL  = ../mips/imin.c
 
+ifdef HAVE_MSA
 SASUMKERNEL  = ../mips/sasum_msa.c
 DASUMKERNEL  = ../mips/dasum_msa.c
 CASUMKERNEL  = ../mips/casum_msa.c
 ZASUMKERNEL  = ../mips/zasum_msa.c
+else
+SASUMKERNEL  = ../mips/asum.c
+DASUMKERNEL  = ../mips/asum.c
+CASUMKERNEL  = ../mips/asum.c
+ZASUMKERNEL  = ../mips/asum.c
+endif
 
 SAXPYKERNEL  = ../mips/axpy.c
 DAXPYKERNEL  = ../mips/axpy.c
@@ -45,10 +52,17 @@ DCOPYKERNEL  = ../mips/copy.c
 CCOPYKERNEL  = ../mips/zcopy.c
 ZCOPYKERNEL  = ../mips/zcopy.c
 
+ifdef HAVE_MSA
 SDOTKERNEL   = ../mips/sdot_msa.c
 DDOTKERNEL   = ../mips/ddot_msa.c
 CDOTKERNEL   = ../mips/cdot_msa.c
 ZDOTKERNEL   = ../mips/zdot_msa.c
+else
+SDOTKERNEL   = ../mips/dot.c
+DDOTKERNEL   = ../mips/dot.c
+CDOTKERNEL   = ../mips/zdot.c
+ZDOTKERNEL   = ../mips/zdot.c
+endif
 
 SNRM2KERNEL  = ../mips/nrm2.c
 DNRM2KERNEL  = ../mips/nrm2.c
@@ -70,22 +84,45 @@ DSWAPKERNEL  = ../mips/swap.c
 CSWAPKERNEL  = ../mips/zswap.c
 ZSWAPKERNEL  = ../mips/zswap.c
 
+ifdef HAVE_MSA
 SGEMVNKERNEL = ../mips/sgemv_n_msa.c
 DGEMVNKERNEL = ../mips/dgemv_n_msa.c
 CGEMVNKERNEL = ../mips/cgemv_n_msa.c
 ZGEMVNKERNEL = ../mips/zgemv_n_msa.c
-
+else
+SGEMVNKERNEL = ../mips/gemv_n.c
+DGEMVNKERNEL = ../mips/gemv_n.c
+CGEMVNKERNEL = ../mips/zgemv_n.c
+ZGEMVNKERNEL = ../mips/zgemv_n.c
+endif
+
+ifdef HAVE_MSA
 SGEMVTKERNEL = ../mips/sgemv_t_msa.c
 DGEMVTKERNEL = ../mips/dgemv_t_msa.c
 CGEMVTKERNEL = ../mips/cgemv_t_msa.c
 ZGEMVTKERNEL = ../mips/zgemv_t_msa.c
-
+else
+SGEMVTKERNEL = ../mips/gemv_t.c
+DGEMVTKERNEL = ../mips/gemv_t.c
+CGEMVTKERNEL = ../mips/zgemv_t.c
+ZGEMVTKERNEL = ../mips/zgemv_t.c
+endif
+
+ifdef HAVE_MSA
 SGEMMKERNEL    = ../mips/sgemm_kernel_8x8_msa.c
 SGEMMONCOPY    = ../mips/sgemm_ncopy_8_msa.c
 SGEMMOTCOPY    = ../mips/sgemm_tcopy_8_msa.c
 SGEMMONCOPYOBJ = sgemm_oncopy.o
 SGEMMOTCOPYOBJ = sgemm_otcopy.o
+else
+SGEMMKERNEL    = ../generic/gemmkernel_2x2.c
+SGEMMONCOPY    = ../generic/gemm_ncopy_2.c
+SGEMMOTCOPY    = ../generic/gemm_tcopy_2.c
+SGEMMONCOPYOBJ = sgemm_oncopy.o
+SGEMMOTCOPYOBJ = sgemm_otcopy.o
+endif
 
+ifdef HAVE_MSA
 DGEMMKERNEL    = ../mips/dgemm_kernel_8x4_msa.c
 DGEMMINCOPY    = ../mips/dgemm_ncopy_8_msa.c
 DGEMMITCOPY    = ../mips/dgemm_tcopy_8_msa.c
@@ -95,7 +132,15 @@ DGEMMINCOPYOBJ = dgemm_incopy.o
 DGEMMITCOPYOBJ = dgemm_itcopy.o
 DGEMMONCOPYOBJ = dgemm_oncopy.o
 DGEMMOTCOPYOBJ = dgemm_otcopy.o
+else
+DGEMMKERNEL    = ../generic/gemmkernel_2x2.c
+DGEMMONCOPY    = ../generic/gemm_ncopy_2.c
+DGEMMOTCOPY    = ../generic/gemm_tcopy_2.c
+DGEMMONCOPYOBJ = dgemm_oncopy.o
+DGEMMOTCOPYOBJ = dgemm_otcopy.o
+endif
 
+ifdef HAVE_MSA
 CGEMMKERNEL    = ../mips/cgemm_kernel_8x4_msa.c
 CGEMMINCOPY    = ../mips/cgemm_ncopy_8_msa.c
 CGEMMITCOPY    = ../mips/cgemm_tcopy_8_msa.c
@@ -105,29 +150,72 @@ CGEMMINCOPYOBJ = cgemm_incopy.o
 CGEMMITCOPYOBJ = cgemm_itcopy.o
 CGEMMONCOPYOBJ = cgemm_oncopy.o
 CGEMMOTCOPYOBJ = cgemm_otcopy.o
+else
+CGEMMKERNEL    = ../generic/zgemmkernel_2x2.c
+CGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
+CGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
+CGEMMONCOPYOBJ = cgemm_oncopy.o
+CGEMMOTCOPYOBJ = cgemm_otcopy.o
+endif
 
+ifdef HAVE_MSA
 ZGEMMKERNEL    = ../mips/zgemm_kernel_4x4_msa.c
 ZGEMMONCOPY    = ../mips/zgemm_ncopy_4_msa.c
 ZGEMMOTCOPY    = ../mips/zgemm_tcopy_4_msa.c
 ZGEMMONCOPYOBJ = zgemm_oncopy.o
 ZGEMMOTCOPYOBJ = zgemm_otcopy.o
+else
+ZGEMMKERNEL    = ../generic/zgemmkernel_2x2.c
+ZGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
+ZGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
+ZGEMMONCOPYOBJ = zgemm_oncopy.o
+ZGEMMOTCOPYOBJ = zgemm_otcopy.o
+endif
 
+ifdef HAVE_MSA
 STRSMKERNEL_LN = ../mips/strsm_kernel_LN_8x8_msa.c
 STRSMKERNEL_LT = ../mips/strsm_kernel_LT_8x8_msa.c
 STRSMKERNEL_RN = ../mips/strsm_kernel_RN_8x8_msa.c
 STRSMKERNEL_RT = ../mips/strsm_kernel_RT_8x8_msa.c
-
+else
+STRSMKERNEL_LN = ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT = ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN = ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT = ../generic/trsm_kernel_RT.c
+endif
+
+ifdef HAVE_MSA
 DTRSMKERNEL_LN = ../mips/dtrsm_kernel_LN_8x4_msa.c
 DTRSMKERNEL_LT = ../mips/dtrsm_kernel_LT_8x4_msa.c
 DTRSMKERNEL_RN = ../mips/dtrsm_kernel_RN_8x4_msa.c
 DTRSMKERNEL_RT = ../mips/dtrsm_kernel_RT_8x4_msa.c
-
+else
+DTRSMKERNEL_LN = ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LT = ../generic/trsm_kernel_LT.c
+DTRSMKERNEL_RN = ../generic/trsm_kernel_RN.c
+DTRSMKERNEL_RT = ../generic/trsm_kernel_RT.c
+endif
+
+ifdef HAVE_MSA
+CTRSMKERNEL_LN = ../generic/trsm_kernel_LN.c
+CTRSMKERNEL_LT = ../generic/trsm_kernel_LT.c
+CTRSMKERNEL_RN = ../generic/trsm_kernel_RN.c
+CTRSMKERNEL_RT = ../generic/trsm_kernel_RT.c
+else
 CTRSMKERNEL_LN = ../generic/trsm_kernel_LN.c
 CTRSMKERNEL_LT = ../generic/trsm_kernel_LT.c
 CTRSMKERNEL_RN = ../generic/trsm_kernel_RN.c
 CTRSMKERNEL_RT = ../generic/trsm_kernel_RT.c
+endif
 
+ifdef HAVE_MSA
+ZTRSMKERNEL_LN = ../generic/trsm_kernel_LN.c
+ZTRSMKERNEL_LT = ../generic/trsm_kernel_LT.c
+ZTRSMKERNEL_RN = ../generic/trsm_kernel_RN.c
+ZTRSMKERNEL_RT = ../generic/trsm_kernel_RT.c
+else
 ZTRSMKERNEL_LN = ../generic/trsm_kernel_LN.c
 ZTRSMKERNEL_LT = ../generic/trsm_kernel_LT.c
 ZTRSMKERNEL_RN = ../generic/trsm_kernel_RN.c
 ZTRSMKERNEL_RT = ../generic/trsm_kernel_RT.c
+endif

param.h

@@ -2174,7 +2174,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SYMV_P	16
 #endif
 
-#if defined(I6400) || defined(P6600)
+#if defined(P5600) || defined(I6400) || defined(P6600)
 #define SNUMOPT  2
 #define DNUMOPT  2
 
@@ -2182,6 +2182,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define GEMM_DEFAULT_OFFSET_B  0
 #define GEMM_DEFAULT_ALIGN  0x03fffUL
 
+#ifdef HAVE_MSA
 #define SGEMM_DEFAULT_UNROLL_M  8
 #define SGEMM_DEFAULT_UNROLL_N  8
 
@@ -2193,46 +2194,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define ZGEMM_DEFAULT_UNROLL_M  4
 #define ZGEMM_DEFAULT_UNROLL_N  4
+#else
+#define SGEMM_DEFAULT_UNROLL_M  2
+#define SGEMM_DEFAULT_UNROLL_N  2
 
-#define SGEMM_DEFAULT_P  128
-#define DGEMM_DEFAULT_P  128
-#define CGEMM_DEFAULT_P  96
-#define ZGEMM_DEFAULT_P  64
-
-#define SGEMM_DEFAULT_Q  240
-#define DGEMM_DEFAULT_Q  120
-#define CGEMM_DEFAULT_Q  120
-#define ZGEMM_DEFAULT_Q  120
-
-#define SGEMM_DEFAULT_R  12288
-#define DGEMM_DEFAULT_R  8192
-#define CGEMM_DEFAULT_R  4096
-#define ZGEMM_DEFAULT_R  4096
+#define DGEMM_DEFAULT_UNROLL_M  2
+#define DGEMM_DEFAULT_UNROLL_N  2
 
+#define CGEMM_DEFAULT_UNROLL_M  2
+#define CGEMM_DEFAULT_UNROLL_N  2
 
-#define SYMV_P  16
+#define ZGEMM_DEFAULT_UNROLL_M  2
+#define ZGEMM_DEFAULT_UNROLL_N  2
 #endif
 
-#if defined(P5600)
-#define SNUMOPT  2
-#define DNUMOPT  2
-
-#define GEMM_DEFAULT_OFFSET_A  0
-#define GEMM_DEFAULT_OFFSET_B  0
-#define GEMM_DEFAULT_ALIGN  0x03fffUL
-
-#define SGEMM_DEFAULT_UNROLL_M  8
-#define SGEMM_DEFAULT_UNROLL_N  8
-
-#define DGEMM_DEFAULT_UNROLL_M  8
-#define DGEMM_DEFAULT_UNROLL_N  4
-
-#define CGEMM_DEFAULT_UNROLL_M  8
-#define CGEMM_DEFAULT_UNROLL_N  4
-
-#define ZGEMM_DEFAULT_UNROLL_M  4
-#define ZGEMM_DEFAULT_UNROLL_N  4
-
 #define SGEMM_DEFAULT_P  128
 #define DGEMM_DEFAULT_P  128
 #define CGEMM_DEFAULT_P  96
@@ -2248,7 +2223,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CGEMM_DEFAULT_R  4096
 #define ZGEMM_DEFAULT_R  4096
 
-
 #define SYMV_P  16
 #endif