Actually add the message to the TLS section

Work around gcc11+ miscompiling C/ZBLAS3 tests at -O3

CMakeLists.txt

@@ -6,7 +6,7 @@ cmake_minimum_required(VERSION 2.8.5)
 project(OpenBLAS C ASM)
 set(OpenBLAS_MAJOR_VERSION 0)
 set(OpenBLAS_MINOR_VERSION 3)
-set(OpenBLAS_PATCH_VERSION 16)
+set(OpenBLAS_PATCH_VERSION 17.dev)
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
 # Adhere to GNU filesystem layout conventions

Changelog.txt

@@ -1,5 +1,20 @@
 OpenBLAS ChangeLog
 ====================================================================
+Version 0.3.17
+ 15-Jul-2021
+
+common:
+ - reverted the optimization of SGEMV_N/DGEMV_N for small input sizes
+   and consecutive arguments as it led to stack overflows on x86_64
+   with some operating systems (notably OSX and Windows)
+ 
+ x86_64:
+  - reverted the performance patch for SGEMV_T on AVX512 as it caused
+    wrong results in some applications
+    
+ SPARC:
+  - fixed compilation with compilers other than gcc
+====================================================================
 Version 0.3.16
  11-Jul-2021
 

Makefile.arm64

@@ -1,4 +1,15 @@
 ifneq ($(C_COMPILER), PGI)
+
+ifneq ($(GCCVERSIONGT4), 1)
+CCOMMON_OPT += -march=armv8-a
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8-a
+endif
+
+
+else 
+
+
 ifeq ($(CORE), ARMV8)
 CCOMMON_OPT += -march=armv8-a
 ifneq ($(F_COMPILER), NAG)
@@ -138,4 +149,7 @@ FCOMMON_OPT += -march=armv8-a -mtune=emag
 endif
 endif
 endif
+
 endif
+
+endif

Makefile.loongarch64

@@ -0,0 +1,3 @@
+ifdef BINARY64
+else
+endif

Makefile.rule

@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.3.16
+VERSION = 0.3.17.dev
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library

Makefile.system

@@ -780,6 +780,11 @@ NO_BINARY_MODE  = 1
 BINARY_DEFINED  = 1
 endif
 
+ifeq ($(ARCH), loongarch64)
+NO_BINARY_MODE  = 1
+BINARY_DEFINED  = 1
+endif
+
 
 #
 #  C Compiler dependent settings
@@ -850,6 +855,13 @@ ifeq ($(OSNAME), AIX)
 BINARY_DEFINED = 1
 endif
 
+ifeq ($(ARCH), loongarch64)
+ifeq ($(CORE), LOONGSONG3R5)
+CCOMMON_OPT += -march=loongarch64 -mabi=lp64
+FCOMMON_OPT += -march=loongarch64 -mabi=lp64
+endif
+endif
+
 endif
 
 ifndef BINARY_DEFINED

TargetList.txt

@@ -110,3 +110,5 @@ Z14
 RISCV64_GENERIC
 C910V
 
+11.LOONGARCH64:
+LOONGSON3R5

c_check

@@ -82,18 +82,19 @@ $os = Interix         if ($data =~ /OS_INTERIX/);
 $os = Android         if ($data =~ /OS_ANDROID/);
 $os = Haiku           if ($data =~ /OS_HAIKU/);
 
-$architecture = x86    if ($data =~ /ARCH_X86/);
-$architecture = x86_64 if ($data =~ /ARCH_X86_64/);
-$architecture = power  if ($data =~ /ARCH_POWER/);
-$architecture = mips   if ($data =~ /ARCH_MIPS/);
-$architecture = mips64 if ($data =~ /ARCH_MIPS64/);
-$architecture = alpha  if ($data =~ /ARCH_ALPHA/);
-$architecture = sparc  if ($data =~ /ARCH_SPARC/);
-$architecture = ia64   if ($data =~ /ARCH_IA64/);
-$architecture = arm    if ($data =~ /ARCH_ARM/);
-$architecture = arm64  if ($data =~ /ARCH_ARM64/);
-$architecture = zarch  if ($data =~ /ARCH_ZARCH/);
-$architecture = riscv64 if ($data =~ /ARCH_RISCV64/);
+$architecture = x86          if ($data =~ /ARCH_X86/);
+$architecture = x86_64       if ($data =~ /ARCH_X86_64/);
+$architecture = power        if ($data =~ /ARCH_POWER/);
+$architecture = mips         if ($data =~ /ARCH_MIPS/);
+$architecture = mips64       if ($data =~ /ARCH_MIPS64/);
+$architecture = alpha        if ($data =~ /ARCH_ALPHA/);
+$architecture = sparc        if ($data =~ /ARCH_SPARC/);
+$architecture = ia64         if ($data =~ /ARCH_IA64/);
+$architecture = arm          if ($data =~ /ARCH_ARM/);
+$architecture = arm64        if ($data =~ /ARCH_ARM64/);
+$architecture = zarch        if ($data =~ /ARCH_ZARCH/);
+$architecture = riscv64      if ($data =~ /ARCH_RISCV64/);
+$architecture = loongarch64  if ($data =~ /ARCH_LOONGARCH64/);
 
 $defined = 0;
 
@@ -143,6 +144,11 @@ if ($architecture eq "riscv64") {
     $binary = 64;
 }
 
+if ($architecture eq "loongarch64") {
+    $defined = 1;
+    $binary = 64;
+}
+
 if ($compiler eq "PGI") {
     $compiler_name .= " -tp p7"    if ($binary eq "32");
     $compiler_name .= " -tp p7-64" if ($binary eq "64");
@@ -215,17 +221,18 @@ if (($architecture eq "mips") || ($architecture eq "mips64")) {
     }
 }
 
-$architecture = x86    if ($data =~ /ARCH_X86/);
-$architecture = x86_64 if ($data =~ /ARCH_X86_64/);
-$architecture = power  if ($data =~ /ARCH_POWER/);
-$architecture = mips   if ($data =~ /ARCH_MIPS/);
-$architecture = mips64 if ($data =~ /ARCH_MIPS64/);
-$architecture = alpha  if ($data =~ /ARCH_ALPHA/);
-$architecture = sparc  if ($data =~ /ARCH_SPARC/);
-$architecture = ia64   if ($data =~ /ARCH_IA64/);
-$architecture = arm    if ($data =~ /ARCH_ARM/);
-$architecture = arm64  if ($data =~ /ARCH_ARM64/);
-$architecture = zarch  if ($data =~ /ARCH_ZARCH/);
+$architecture = x86          if ($data =~ /ARCH_X86/);
+$architecture = x86_64       if ($data =~ /ARCH_X86_64/);
+$architecture = power        if ($data =~ /ARCH_POWER/);
+$architecture = mips         if ($data =~ /ARCH_MIPS/);
+$architecture = mips64       if ($data =~ /ARCH_MIPS64/);
+$architecture = alpha        if ($data =~ /ARCH_ALPHA/);
+$architecture = sparc        if ($data =~ /ARCH_SPARC/);
+$architecture = ia64         if ($data =~ /ARCH_IA64/);
+$architecture = arm          if ($data =~ /ARCH_ARM/);
+$architecture = arm64        if ($data =~ /ARCH_ARM64/);
+$architecture = zarch        if ($data =~ /ARCH_ZARCH/);
+$architecture = loongarch64  if ($data =~ /ARCH_LOONGARCH64/);
 
 $binformat    = bin32;
 $binformat    = bin64  if ($data =~ /BINARY_64/);

cmake/cc.cmake

@@ -124,9 +124,9 @@ if (NOT DYNAMIC_ARCH)
 	if (HAVE_AVX)
         set (CCOMMON_OPT  "${CCOMMON_OPT} -mavx")
 	endif ()
-	if (HAVE_FMA3)
-	set (CCOMMON_OPT  "${CCOMMON_OPT} -mfma")
-	endif ()
+	#	if (HAVE_FMA3)
+	#set (CCOMMON_OPT  "${CCOMMON_OPT} -mfma")
+	#endif ()
 	if (HAVE_SSE)
 	set (CCOMMON_OPT  "${CCOMMON_OPT} -msse")
 	endif ()

cmake/system.cmake

@@ -186,11 +186,11 @@ if (DEFINED TARGET)
       	  set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mavx2")
 	endif()
   endif()
-  if (DEFINED HAVE_FMA3)
-	if (NOT NO_AVX2)
-    set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mfma")
-	endif()
-  endif()
+  #  if (DEFINED HAVE_FMA3)
+  #	if (NOT NO_AVX2)
+  #  set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mfma")
+  #	endif()
+  #  endif()
     if (DEFINED HAVE_SSE)
     set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse")
   endif()

cmake/utils.cmake

@@ -311,7 +311,15 @@ function(GenerateNamedObjects sources_in)
       configure_file(${new_source_file}.tmp ${new_source_file} COPYONLY)
       file(REMOVE ${new_source_file}.tmp)
       list(APPEND SRC_LIST_OUT ${new_source_file})
-
+      message (STATUS ${new_source_file})
+      if (DEFINED HAVE_FMA3)
+        if ( ${new_source_file} MATCHES "(s|d?)rot_k.*c")
+		set_source_files_properties(${new_source_file} PROPERTIES COMPILE_OPTIONS "-mfma")
+        endif ()
+        if ( ${new_source_file} MATCHES "dgemv_t_k.*c")
+		set_source_files_properties(${new_source_file} PROPERTIES COMPILE_OPTIONS "-mfma")
+        endif ()
+      endif ()
     endforeach ()
   endforeach ()
 

common.h

@@ -449,7 +449,7 @@ please https://github.com/xianyi/OpenBLAS/issues/246
 #include "common_mips.h"
 #endif
 
-    
+
 #ifdef ARCH_RISCV64
 #include "common_riscv64.h"
 #endif
@@ -470,6 +470,10 @@ please https://github.com/xianyi/OpenBLAS/issues/246
 #include "common_zarch.h"
 #endif
 
+#ifdef ARCH_LOONGARCH64
+#include "common_loongarch64.h"
+#endif
+
 #ifndef ASSEMBLER
 #ifdef OS_WINDOWSSTORE
 typedef char env_var_t[MAX_PATH];

common_loongarch64.h

@@ -0,0 +1,199 @@
+/*****************************************************************************
+Copyright (c) 2011-2020, The OpenBLAS Project
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+   1. Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in
+      the documentation and/or other materials provided with the
+      distribution.
+   3. Neither the name of the OpenBLAS project nor the names of
+      its contributors may be used to endorse or promote products
+      derived from this software without specific prior written
+      permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+**********************************************************************************/
+
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#ifndef COMMON_LOONGARCH64
+#define COMMON_LOONGARCH64
+
+#define MB  __sync_synchronize()
+#define WMB __sync_synchronize()
+#define RMB __sync_synchronize()
+
+#define INLINE inline
+
+#ifndef ASSEMBLER
+
+static inline int blas_quickdivide(blasint x, blasint y){
+  return x / y;
+}
+
+#ifdef DOUBLE
+#define GET_IMAGE(res)  __asm__ __volatile__("fmov.d %0, $f2" : "=f"(res)  : : "memory")
+#else
+#define GET_IMAGE(res)  __asm__ __volatile__("fmov.s %0, $f2" : "=f"(res)  : : "memory")
+#endif
+
+#define GET_IMAGE_CANCEL
+
+#else
+
+#ifdef DOUBLE
+#define LD      fld.d
+#define ST      fst.d
+#define MADD    fmadd.d
+#define NMADD   fnmadd.d
+#define MSUB    fmsub.d
+#define NMSUB   fnmsub.d
+#define ADD     fadd.d
+#define SUB     fsub.d
+#define MUL     fmul.d
+#define MOV     fmov.d
+#define CMOVT   fsel
+#define MTC     movgr2fr.d
+#define FABS    fabs.d
+#define CMPEQ   fcmp.ceq.d
+#define CMPLE   fcmp.cle.d
+#define CMPLT   fcmp.clt.d
+#define NEG     fneg.d
+#else
+#define LD      fld.s
+#define ST      fst.s
+#define MADD    fmadd.s
+#define NMADD   fnmadd.s
+#define MSUB    fmsub.s
+#define NMSUB   fnmsub.s
+#define ADD     fadd.s
+#define SUB     fsub.s
+#define MUL     fmul.s
+#define MOV     fmov.s
+#define CMOVT   fsel
+#define MTC     movgr2fr.w
+#define FABS    fabs.s
+#define CMPEQ   fcmp.ceq.s
+#define CMPLE   fcmp.cle.s
+#define CMPLT   fcmp.clt.s
+#define NEG     fneg.s
+#endif /* defined(DOUBLE) */
+
+#if defined(__64BIT__) && defined(USE64BITINT)
+#define LDINT   ld.d
+#define LDARG   ld.d
+#define SDARG   st.d
+#elif defined(__64BIT__) && !defined(USE64BITINT)
+#define LDINT   ld.w
+#define LDARG   ld.d
+#define SDARG   st.d
+#else
+#define LDINT   ld.w
+#define LDARG   ld.w
+#define SDARG   st.w
+#endif
+
+
+#ifndef F_INTERFACE
+#define REALNAME ASMNAME
+#else
+#define REALNAME ASMFNAME
+#endif /* defined(F_INTERFACE) */
+
+#if defined(ASSEMBLER) && !defined(NEEDPARAM)
+
+#define PROLOGUE \
+    .text ;\
+    .align 5 ;\
+    .globl  REALNAME ;\
+    .type   REALNAME, @function ;\
+REALNAME: ;\
+
+#if defined(__linux__) && defined(__ELF__)
+#define GNUSTACK .section .note.GNU-stack,"",@progbits
+#else
+#define GNUSTACK
+#endif /* defined(__linux__) && defined(__ELF__) */
+
+#define EPILOGUE      \
+    .end    REALNAME ;\
+    GNUSTACK
+
+#define PROFCODE
+
+#define MOVT(dst, src, cc)  \
+    bceqz  cc,   1f;        \
+    add.d dst,  src,  $r0;  \
+    1:
+
+#endif /* defined(ASSEMBLER) && !defined(NEEDPARAM) */
+
+#endif /* defined(ASSEMBLER) */
+
+#define SEEK_ADDRESS
+
+#define BUFFER_SIZE     ( 32 << 20)
+
+#define PAGESIZE        (16UL << 1)
+#define FIXED_PAGESIZE  (16UL << 10)
+#define HUGE_PAGESIZE   ( 2 << 20)
+
+#define BASE_ADDRESS (START_ADDRESS - BUFFER_SIZE * MAX_CPU_NUMBER)
+
+#ifndef MAP_ANONYMOUS
+#define MAP_ANONYMOUS MAP_ANON
+#endif
+
+#endif

common_macro.h

@@ -2490,7 +2490,8 @@
 #endif
 
 #ifndef ASSEMBLER
-#if defined(ARCH_X86) || defined(ARCH_X86_64) || defined(ARCH_IA64) || defined(ARCH_MIPS64) || defined(ARCH_ARM64)
+#if defined(ARCH_X86) || defined(ARCH_X86_64) || defined(ARCH_IA64) || defined(ARCH_MIPS64) || defined(ARCH_ARM64)\
+|| defined(ARCH_LOONGARCH64)
 extern BLASLONG gemm_offset_a;
 extern BLASLONG gemm_offset_b;
 extern BLASLONG sbgemm_p;

cpuid.h

@@ -54,6 +54,7 @@
 #define VENDOR_TRANSMETA  9
 #define VENDOR_NSC	 10
 #define VENDOR_HYGON	 11
+#define VENDOR_ZHAOXIN   12
 #define VENDOR_UNKNOWN   99
 
 #define BITMASK(a, b, c) ((((a) >> (b)) & (c)))

cpuid_loongarch64.c

@@ -0,0 +1,110 @@
+/*****************************************************************************
+Copyright (c) 2011-2020, The OpenBLAS Project
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+   1. Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in
+      the documentation and/or other materials provided with the
+      distribution.
+   3. Neither the name of the OpenBLAS project nor the names of
+      its contributors may be used to endorse or promote products
+      derived from this software without specific prior written
+      permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+**********************************************************************************/
+
+#include <stdint.h>
+
+#define CPU_UNKNOWN     0
+#define CPU_LOONGSON3R5 1
+
+#define LOONGARCH_CFG2  0x02
+#define LOONGARCH_LASX  1<<7
+
+static char *cpuname[] = {
+  "UNKNOWN",
+  "LOONGSON3R5"
+};
+
+int detect(void) {
+    uint32_t reg = 0;
+
+    __asm__ volatile (
+        "cpucfg %0, %1 \n\t"
+        : "+&r"(reg)
+        : "r"(LOONGARCH_CFG2)
+    );
+
+    if (reg & LOONGARCH_LASX)
+        return CPU_LOONGSON3R5;
+    else
+        return CPU_UNKNOWN;
+}
+
+char *get_corename(void) {
+  return cpuname[detect()];
+}
+
+void get_architecture(void) {
+  printf("LOONGARCH64");
+}
+
+void get_subarchitecture(void) {
+  if (detect() == CPU_LOONGSON3R5) {
+    printf("LOONGSON3R5");
+  } else {
+    printf("UNKNOWN");
+  }
+}
+
+void get_subdirname(void) {
+  printf("loongarch64");
+}
+
+void get_cpuconfig(void) {
+  if (detect() == CPU_LOONGSON3R5) {
+    printf("#define LOONGSON3R5\n");
+    printf("#define L1_DATA_SIZE 65536\n");
+    printf("#define L1_DATA_LINESIZE 64\n");
+    printf("#define L2_SIZE 1048576\n");
+    printf("#define L2_LINESIZE 64\n");
+    printf("#define DTB_DEFAULT_ENTRIES 64\n");
+    printf("#define DTB_SIZE 4096\n");
+    printf("#define L2_ASSOCIATIVE 16\n");
+  } else {
+    printf("#define LOONGSON3R5\n");
+    printf("#define L1_DATA_SIZE 65536\n");
+    printf("#define L1_DATA_LINESIZE 64\n");
+    printf("#define L2_SIZE 1048576\n");
+    printf("#define L2_LINESIZE 64\n");
+    printf("#define DTB_DEFAULT_ENTRIES 64\n");
+    printf("#define DTB_SIZE 4096\n");
+    printf("#define L2_ASSOCIATIVE 16\n");
+  }
+}
+
+void get_libname(void){
+  if (detect() == CPU_LOONGSON3R5) {
+    printf("loongson3r5\n");
+  } else {
+    printf("loongarch64\n");
+  }
+}

cpuid_x86.c

@@ -283,7 +283,7 @@ int get_vendor(void){
   if (!strcmp(vendor, "CyrixInstead")) return VENDOR_CYRIX;
   if (!strcmp(vendor, "NexGenDriven")) return VENDOR_NEXGEN;
   if (!strcmp(vendor, "CentaurHauls")) return VENDOR_CENTAUR;
-  if (!strcmp(vendor, "  Shanghai  ")) return VENDOR_CENTAUR;
+  if (!strcmp(vendor, "  Shanghai  ")) return VENDOR_ZHAOXIN;
   if (!strcmp(vendor, "RiseRiseRise")) return VENDOR_RISE;
   if (!strcmp(vendor, " SiS SiS SiS")) return VENDOR_SIS;
   if (!strcmp(vendor, "GenuineTMx86")) return VENDOR_TRANSMETA;
@@ -1067,7 +1067,8 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
 
   if ((get_vendor() == VENDOR_AMD) ||
       (get_vendor() == VENDOR_HYGON) ||
-      (get_vendor() == VENDOR_CENTAUR)) {
+      (get_vendor() == VENDOR_CENTAUR) ||
+      (get_vendor() == VENDOR_ZHAOXIN)) {
     cpuid(0x80000005, &eax, &ebx, &ecx, &edx);
 
     LDTB.size        = 4096;
@@ -1190,7 +1191,7 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
 
 int get_cpuname(void){
 
-  int family, exfamily, model, vendor, exmodel;
+  int family, exfamily, model, vendor, exmodel, stepping;
 
   if (!have_cpuid()) return CPUTYPE_80386;
 
@@ -1198,6 +1199,7 @@ int get_cpuname(void){
   exfamily = get_cputype(GET_EXFAMILY);
   model    = get_cputype(GET_MODEL);
   exmodel  = get_cputype(GET_EXMODEL);
+  stepping = get_cputype(GET_STEPPING);
 
   vendor = get_vendor();
 
@@ -1628,15 +1630,20 @@ int get_cpuname(void){
     switch (family) {
     case 0x5:
       return CPUTYPE_CENTAURC6;
-      break;
     case 0x6:
-      return CPUTYPE_NANO;
-      break;
-    case 0x7:
+      if (model == 0xf && stepping < 0xe)
+        return CPUTYPE_NANO;
       return CPUTYPE_NEHALEM;
-      break;
+    default:
+      if (family >= 0x7)
+        return CPUTYPE_NEHALEM;
+      else
+        return CPUTYPE_VIAC3;
     }
-    return CPUTYPE_VIAC3;
+  }
+
+  if (vendor == VENDOR_ZHAOXIN){
+    return CPUTYPE_NEHALEM;
   }
 
   if (vendor == VENDOR_RISE){
@@ -1869,7 +1876,7 @@ char *get_lower_cpunamechar(void){
 
 int get_coretype(void){
 
-  int family, exfamily, model, exmodel, vendor;
+  int family, exfamily, model, exmodel, vendor, stepping;
 
   if (!have_cpuid()) return CORE_80486;
 
@@ -1877,6 +1884,7 @@ int get_coretype(void){
   exfamily = get_cputype(GET_EXFAMILY);
   model    = get_cputype(GET_MODEL);
   exmodel  = get_cputype(GET_EXMODEL);
+  stepping = get_cputype(GET_STEPPING);
 
   vendor = get_vendor();
 
@@ -2286,13 +2294,19 @@ int get_coretype(void){
   if (vendor == VENDOR_CENTAUR) {
     switch (family) {
     case 0x6:
-      return CORE_NANO;
-      break;
-    case 0x7:
+      if (model == 0xf && stepping < 0xe)
+        return CORE_NANO;
       return CORE_NEHALEM;
-      break;
+    default:
+      if (family >= 0x7)
+        return CORE_NEHALEM;
+      else
+        return CORE_VIAC3;
     }
-    return CORE_VIAC3;
+  }
+
+  if (vendor == VENDOR_ZHAOXIN) {
+     return CORE_NEHALEM;
   }
 
   return CORE_UNKNOWN;

ctest.c

@@ -157,6 +157,10 @@ ARCH_ARM64
 ARCH_RISCV64
 #endif
 
+#ifdef __loongarch64
+ARCH_LOONGARCH64
+#endif
+
 #if (defined(__STDC_VERSION__) && __STDC_VERSION__ >= 201112L)
 HAVE_C11
 #endif

ctest/CMakeLists.txt

@@ -4,6 +4,9 @@ include_directories(${PROJECT_BINARY_DIR})
 enable_language(Fortran)
 
 set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DADD${BU} -DCBLAS")
+if (CMAKE_Fortran_COMPILER_ID STREQUAL GNU)
+	set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -fno-tree-vectorize")
+endif()
 
 if(WIN32)
 FILE(WRITE ${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.ps1

ctest/Makefile

@@ -6,6 +6,9 @@ TOPDIR = ..
 include $(TOPDIR)/Makefile.system
 
 override CFLAGS += -DADD$(BU) -DCBLAS
+ifeq ($(F_COMPILER),GFORTRAN)
+	override FFLAGS += -fno-tree-vectorize
+endif
 override TARGET_ARCH=
 override TARGET_MACH=
 

driver/others/dynamic.c

@@ -292,6 +292,7 @@ extern gotoblas_t  gotoblas_COOPERLAKE;
 #define VENDOR_AMD        2
 #define VENDOR_CENTAUR    3
 #define VENDOR_HYGON	  4
+#define VENDOR_ZHAOXIN    5
 #define VENDOR_UNKNOWN   99
 
 #define BITMASK(a, b, c) ((((a) >> (b)) & (c)))
@@ -404,7 +405,7 @@ static int get_vendor(void){
   if (!strcmp(vendor.vchar, "GenuineIntel")) return VENDOR_INTEL;
   if (!strcmp(vendor.vchar, "AuthenticAMD")) return VENDOR_AMD;
   if (!strcmp(vendor.vchar, "CentaurHauls")) return VENDOR_CENTAUR;
-  if (!strcmp(vendor.vchar, "  Shanghai  ")) return VENDOR_CENTAUR;
+  if (!strcmp(vendor.vchar, "  Shanghai  ")) return VENDOR_ZHAOXIN;
   if (!strcmp(vendor.vchar, "HygonGenuine")) return VENDOR_HYGON;
 
   if ((eax == 0) || ((eax & 0x500) != 0)) return VENDOR_INTEL;
@@ -415,7 +416,7 @@ static int get_vendor(void){
 static gotoblas_t *get_coretype(void){
 
   int eax, ebx, ecx, edx;
-  int family, exfamily, model, vendor, exmodel;
+  int family, exfamily, model, vendor, exmodel, stepping;
 
   cpuid(1, &eax, &ebx, &ecx, &edx);
 
@@ -423,6 +424,7 @@ static gotoblas_t *get_coretype(void){
   exfamily = BITMASK(eax, 20, 0xff);
   model    = BITMASK(eax,  4, 0x0f);
   exmodel  = BITMASK(eax, 16, 0x0f);
+  stepping = BITMASK(eax,  0, 0x0f);
 
   vendor = get_vendor();
 
@@ -824,13 +826,19 @@ static gotoblas_t *get_coretype(void){
   if (vendor == VENDOR_CENTAUR) {
     switch (family) {
     case 0x6:
-      return &gotoblas_NANO;
-      break;
-    case 0x7:
+      if (model == 0xf && stepping < 0xe)
+        return &gotoblas_NANO;
       return &gotoblas_NEHALEM;
+    default:
+      if (family >= 0x7)
+        return &gotoblas_NEHALEM;
     }
   }
 
+  if (vendor == VENDOR_ZHAOXIN) {
+      return &gotoblas_NEHALEM;
+  }
+
   return NULL;
 }
 

driver/others/memory.c

@@ -428,7 +428,7 @@ extern int openblas_goto_num_threads_env();
 extern int openblas_omp_num_threads_env();
 
 int blas_get_cpu_number(void){
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
   int max_num;
 #endif
   int blas_goto_num   = 0;
@@ -436,7 +436,7 @@ int blas_get_cpu_number(void){
 
   if (blas_num_threads) return blas_num_threads;
 
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
   max_num = get_num_procs();
 #endif
 
@@ -460,7 +460,7 @@ int blas_get_cpu_number(void){
   else if (blas_omp_num > 0) blas_num_threads = blas_omp_num;
   else blas_num_threads = MAX_CPU_NUMBER;
 
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
   if (blas_num_threads > max_num) blas_num_threads = max_num;
 #endif
 
@@ -1291,7 +1291,12 @@ UNLOCK_COMMAND(&alloc_lock);
   return (void *)(((char *)alloc_info) + sizeof(struct alloc_t));
 
  error:
-  printf("OpenBLAS : Program will terminate because you tried to allocate too many memory regions.\n");
+  printf("OpenBLAS : Program will terminate because you tried to allocate too many TLS memory regions.\n");
+  printf("This library was built to support a maximum of %d threads - either rebuild OpenBLAS\n", NUM_BUFFERS);
+  printf("with a larger NUM_THREADS value or set the environment variable OPENBLAS_NUM_THREADS to\n");
+  printf("a sufficiently small number. This error typically occurs when the software that relies on\n");
+  printf("OpenBLAS calls BLAS functions from many threads in parallel, or when your computer has more\n");
+  printf("cpu cores than what OpenBLAS was configured to handle.\n"); 
 
   return NULL;
 }
@@ -1979,7 +1984,7 @@ extern int openblas_goto_num_threads_env();
 extern int openblas_omp_num_threads_env();
 
 int blas_get_cpu_number(void){
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
   int max_num;
 #endif
   int blas_goto_num   = 0;
@@ -1987,7 +1992,7 @@ int blas_get_cpu_number(void){
 
   if (blas_num_threads) return blas_num_threads;
 
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
   max_num = get_num_procs();
 #endif
 
@@ -2011,7 +2016,7 @@ int blas_get_cpu_number(void){
   else if (blas_omp_num > 0) blas_num_threads = blas_omp_num;
   else blas_num_threads = MAX_CPU_NUMBER;
 
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
   if (blas_num_threads > max_num) blas_num_threads = max_num;
 #endif
 
@@ -2878,8 +2883,12 @@ void *blas_memory_alloc(int procpos){
   return (void *)memory[position].addr;
 
  error:
-  printf("BLAS : Program is Terminated. Because you tried to allocate too many memory regions.\n");
-
+  printf("OpenBLAS : Program is Terminated. Because you tried to allocate too many memory regions.\n");
+  printf("This library was built to support a maximum of %d threads - either rebuild OpenBLAS\n", NUM_BUFFERS);
+  printf("with a larger NUM_THREADS value or set the environment variable OPENBLAS_NUM_THREADS to\n");
+  printf("a sufficiently small number. This error typically occurs when the software that relies on\n");
+  printf("OpenBLAS calls BLAS functions from many threads in parallel, or when your computer has more\n");
+  printf("cpu cores than what OpenBLAS was configured to handle.\n"); 
   return NULL;
 }
 

getarch.c

@@ -142,6 +142,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /* #define FORCE_SICORTEX	*/
 /* #define FORCE_LOONGSON3R3	*/
 /* #define FORCE_LOONGSON3R4	*/
+/* #define FORCE_LOONGSON3R5	*/
 /* #define FORCE_I6400		*/
 /* #define FORCE_P6600		*/
 /* #define FORCE_P5600		*/
@@ -842,6 +843,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #else
 #endif
 
+#ifdef FORCE_LOONGSON3R5
+#define FORCE
+#define ARCHITECTURE    "LOONGARCH"
+#define SUBARCHITECTURE "LOONGSON3R5"
+#define SUBDIRNAME      "loongarch64"
+#define ARCHCONFIG   "-DLOONGSON3R5 " \
+       "-DL1_DATA_SIZE=65536 -DL1_DATA_LINESIZE=64 " \
+       "-DL2_SIZE=1048576 -DL2_LINESIZE=64 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=16 "
+#define LIBNAME   "loongson3r5"
+#define CORENAME  "LOONGSON3R5"
+#else
+#endif
+
 #ifdef FORCE_I6400
 #define FORCE
 #define ARCHITECTURE    "MIPS"
@@ -1388,6 +1403,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define OPENBLAS_SUPPORTED
 #endif
 
+#ifdef __loongarch64
+#include "cpuid_loongarch64.c"
+#define OPENBLAS_SUPPORTED
+#endif
+
 #ifdef __riscv
 #include "cpuid_riscv64.c"
 #define OPENBLAS_SUPPORTED
@@ -1463,7 +1483,7 @@ int main(int argc, char *argv[]){
 #ifdef FORCE
     printf("CORE=%s\n", CORENAME);
 #else
-#if defined(INTEL_AMD) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__) || defined(ZARCH) || defined(sparc)
+#if defined(INTEL_AMD) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__) || defined(ZARCH) || defined(sparc) || defined(__loongarch__)
     printf("CORE=%s\n", get_corename());
 #endif
 #endif
@@ -1611,7 +1631,7 @@ printf("ELF_VERSION=2\n");
 #ifdef FORCE
     printf("#define CHAR_CORENAME \"%s\"\n", CORENAME);
 #else
-#if defined(INTEL_AMD) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__) || defined(ZARCH) || defined(sparc)
+#if defined(INTEL_AMD) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__) || defined(ZARCH) || defined(sparc) || defined(__loongarch__)
     printf("#define CHAR_CORENAME \"%s\"\n", get_corename());
 #endif
 #endif

interface/gemv.c

@@ -201,12 +201,14 @@ void CNAME(enum CBLAS_ORDER order,
   if (beta != ONE) SCAL_K(leny, 0, 0, beta, y, blasabs(incy), NULL, 0, NULL, 0);
 
   if (alpha == ZERO) return;
-
+	
+#if 0
+/* this optimization causes stack corruption on x86_64 under OSX, Windows and FreeBSD */	
   if (trans == 0 && incx == 1 && incy == 1 && m*n < 2304 *GEMM_MULTITHREAD_THRESHOLD) {
     GEMV_N(m, n, 0, alpha, a, lda, x, incx, y, incy, NULL);
     return;
   }    
-
+#endif
   IDEBUG_START;
 
   FUNCTION_PROFILE_START();

kernel/Makefile.L2

@@ -1,3 +1,10 @@
+FMAFLAG=
+ifndef OLDGCC
+ifdef HAVE_FMA3
+FMAFLAG = -mfma
+endif
+endif
+
 ### GEMV ###
 
 ifndef SGEMVNKERNEL
@@ -263,7 +270,7 @@ $(KDIR)dgemv_n$(TSUFFIX).$(SUFFIX)  $(KDIR)dgemv_n$(TSUFFIX).$(PSUFFIX)  : $(KER
 	$(CC) -c $(CFLAGS) -DDOUBLE -UCOMPLEX  -UTRANS $< -o $@
 
 $(KDIR)dgemv_t$(TSUFFIX).$(SUFFIX)  $(KDIR)dgemv_t$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DGEMVTKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -DDOUBLE -UCOMPLEX  -DTRANS $< -o $@
+	$(CC) -c $(CFLAGS) $(FMAFLAG) -DDOUBLE -UCOMPLEX  -DTRANS $< -o $@
 endif
 
 $(KDIR)qgemv_n$(TSUFFIX).$(SUFFIX)  $(KDIR)qgemv_n$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QGEMVNKERNEL)

kernel/loongarch64/KERNEL

@@ -0,0 +1,236 @@
+ifndef SAXPYKERNEL
+SAXPYKERNEL = ../arm/axpy.c
+endif
+
+ifndef DAXPYKERNEL
+DAXPYKERNEL = ../arm/axpy.c
+endif
+
+ifndef CAXPYKERNEL
+CAXPYKERNEL = ../arm/zaxpy.c
+endif
+
+ifndef ZAXPYKERNEL
+ZAXPYKERNEL = ../arm/zaxpy.c
+endif
+
+ifndef SROTKERNEL
+SROTKERNEL  = ../arm/rot.c
+endif
+
+ifndef DROTKERNEL
+DROTKERNEL  = ../arm/rot.c
+endif
+
+ifndef CROTKERNEL
+CROTKERNEL  = ../arm/zrot.c
+endif
+
+ifndef ZROTKERNEL
+ZROTKERNEL  = ../arm/zrot.c
+endif
+
+ifndef CSWAPKERNEL
+CSWAPKERNEL = ../arm/zswap.c
+endif
+
+ifndef ZSWAPKERNEL
+ZSWAPKERNEL = ../arm/zswap.c
+endif
+
+ifndef SSUMKERNEL
+SSUMKERNEL  = ../arm/sum.c
+endif
+
+ifndef DSUMKERNEL
+DSUMKERNEL  = ../arm/sum.c
+endif
+
+ifndef CSUMKERNEL
+CSUMKERNEL  = ../arm/zsum.c
+endif
+
+ifndef ZSUMKERNEL
+ZSUMKERNEL  = ../arm/zsum.c
+endif
+
+ifndef ISMAXKERNEL
+ISMAXKERNEL = ../arm/imax.c
+endif
+
+ifndef IDMAXKERNEL
+IDMAXKERNEL = ../arm/imax.c
+endif
+
+ifndef ISMINKERNEL
+ISMINKERNEL = ../arm/imin.c
+endif
+
+ifndef IDMINKERNEL
+IDMINKERNEL = ../arm/imin.c
+endif
+
+ifndef SNRM2KERNEL
+SNRM2KERNEL = snrm2.S
+endif
+
+ifndef DNRM2KERNEL
+DNRM2KERNEL = dnrm2.S
+endif
+
+ifndef CNRM2KERNEL
+CNRM2KERNEL = cnrm2.S
+endif
+
+ifndef ZNRM2KERNEL
+ZNRM2KERNEL = znrm2.S
+endif
+
+ifndef SCABS_KERNEL
+SCABS_KERNEL   = ../generic/cabs.c
+endif
+
+ifndef DCABS_KERNEL
+DCABS_KERNEL   = ../generic/cabs.c
+endif
+
+ifndef QCABS_KERNEL
+QCABS_KERNEL   = ../generic/cabs.c
+endif
+
+ifndef LSAME_KERNEL
+LSAME_KERNEL   = ../generic/lsame.c
+endif
+
+ifndef SGEMMKERNEL
+SGEMMKERNEL    =  gemm_kernel.S
+SGEMMINCOPY    = ../generic/gemm_ncopy_2.c
+SGEMMITCOPY    = ../generic/gemm_tcopy_2.c
+SGEMMONCOPY    = ../generic/gemm_ncopy_8.c
+SGEMMOTCOPY    = ../generic/gemm_tcopy_8.c
+SGEMMINCOPYOBJ =  sgemm_incopy.o
+SGEMMITCOPYOBJ =  sgemm_itcopy.o
+SGEMMONCOPYOBJ =  sgemm_oncopy.o
+SGEMMOTCOPYOBJ =  sgemm_otcopy.o
+endif
+
+ifndef DGEMMKERNEL
+DGEMMKERNEL    =  gemm_kernel.S
+DGEMMINCOPY    = ../generic/gemm_ncopy_2.c
+DGEMMITCOPY    = ../generic/gemm_tcopy_2.c
+DGEMMONCOPY    = ../generic/gemm_ncopy_8.c
+DGEMMOTCOPY    = ../generic/gemm_tcopy_8.c
+DGEMMINCOPYOBJ =  dgemm_incopy.o
+DGEMMITCOPYOBJ =  dgemm_itcopy.o
+DGEMMONCOPYOBJ =  dgemm_oncopy.o
+DGEMMOTCOPYOBJ =  dgemm_otcopy.o
+endif
+
+ifndef CGEMMKERNEL
+CGEMMKERNEL    =  zgemm_kernel.S
+CGEMMINCOPY    = ../generic/zgemm_ncopy_1.c
+CGEMMITCOPY    = ../generic/zgemm_tcopy_1.c
+CGEMMONCOPY    = ../generic/zgemm_ncopy_4.c
+CGEMMOTCOPY    = ../generic/zgemm_tcopy_4.c
+CGEMMINCOPYOBJ =  cgemm_incopy.o
+CGEMMITCOPYOBJ =  cgemm_itcopy.o
+CGEMMONCOPYOBJ =  cgemm_oncopy.o
+CGEMMOTCOPYOBJ =  cgemm_otcopy.o
+endif
+
+ifndef ZGEMMKERNEL
+ZGEMMKERNEL    =  zgemm_kernel.S
+ZGEMMINCOPY    = ../generic/zgemm_ncopy_1.c
+ZGEMMITCOPY    = ../generic/zgemm_tcopy_1.c
+ZGEMMONCOPY    = ../generic/zgemm_ncopy_4.c
+ZGEMMOTCOPY    = ../generic/zgemm_tcopy_4.c
+ZGEMMINCOPYOBJ =  zgemm_incopy.o
+ZGEMMITCOPYOBJ =  zgemm_itcopy.o
+ZGEMMONCOPYOBJ =  zgemm_oncopy.o
+ZGEMMOTCOPYOBJ =  zgemm_otcopy.o
+endif
+
+ifndef SGEMM_BETA
+SGEMM_BETA = ../generic/gemm_beta.c
+endif
+ifndef DGEMM_BETA
+DGEMM_BETA = ../generic/gemm_beta.c
+endif
+ifndef CGEMM_BETA
+CGEMM_BETA = ../generic/zgemm_beta.c
+endif
+ifndef ZGEMM_BETA
+ZGEMM_BETA = ../generic/zgemm_beta.c
+endif
+
+ifndef STRSMKERNEL_LN
+STRSMKERNEL_LN =  trsm_kernel_LN.S
+endif
+
+ifndef STRSMKERNEL_LT
+STRSMKERNEL_LT =  trsm_kernel_LT.S
+endif
+
+ifndef STRSMKERNEL_RN
+STRSMKERNEL_RN =  trsm_kernel_LT.S
+endif
+
+ifndef STRSMKERNEL_RT
+STRSMKERNEL_RT =  trsm_kernel_RT.S
+endif
+
+ifndef DTRSMKERNEL_LN
+DTRSMKERNEL_LN =  trsm_kernel_LN.S
+endif
+
+ifndef DTRSMKERNEL_LT
+DTRSMKERNEL_LT =  trsm_kernel_LT.S
+endif
+
+ifndef DTRSMKERNEL_RN
+DTRSMKERNEL_RN =  trsm_kernel_LT.S
+endif
+
+ifndef DTRSMKERNEL_RT
+DTRSMKERNEL_RT =  trsm_kernel_RT.S
+endif
+
+ifndef CTRSMKERNEL_LN
+CTRSMKERNEL_LN =  ztrsm_kernel_LT.S
+endif
+
+ifndef CTRSMKERNEL_LT
+CTRSMKERNEL_LT =  ztrsm_kernel_LT.S
+endif
+
+ifndef CTRSMKERNEL_RN
+CTRSMKERNEL_RN =  ztrsm_kernel_LT.S
+endif
+
+ifndef CTRSMKERNEL_RT
+CTRSMKERNEL_RT =  ztrsm_kernel_RT.S
+endif
+
+ifndef ZTRSMKERNEL_LN
+ZTRSMKERNEL_LN =  ztrsm_kernel_LT.S
+endif
+
+ifndef ZTRSMKERNEL_LT
+ZTRSMKERNEL_LT =  ztrsm_kernel_LT.S
+endif
+
+ifndef ZTRSMKERNEL_RN
+ZTRSMKERNEL_RN =  ztrsm_kernel_LT.S
+endif
+
+ifndef ZTRSMKERNEL_RT
+ZTRSMKERNEL_RT =  ztrsm_kernel_RT.S
+endif
+
+ifndef CGEMM3MKERNEL
+CGEMM3MKERNEL    =  zgemm3m_kernel.S
+endif
+
+ifndef ZGEMM3MKERNEL
+ZGEMM3MKERNEL    =  zgemm3m_kernel.S
+endif

kernel/loongarch64/KERNEL.LOONGSON3R5

@@ -0,0 +1 @@
+#TODO: Add loongarch64 SIMD optimizations

kernel/loongarch64/KERNEL.generic

@@ -0,0 +1,167 @@
+SGEMM_BETA = ../generic/gemm_beta.c
+DGEMM_BETA = ../generic/gemm_beta.c
+CGEMM_BETA = ../generic/zgemm_beta.c
+ZGEMM_BETA = ../generic/zgemm_beta.c
+
+STRMMKERNEL    = ../generic/trmmkernel_2x2.c
+DTRMMKERNEL    = ../generic/trmmkernel_2x2.c
+CTRMMKERNEL    = ../generic/ztrmmkernel_2x2.c
+ZTRMMKERNEL    = ../generic/ztrmmkernel_2x2.c
+
+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
+
+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
+
+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
+
+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
+
+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
+
+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
+
+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
+
+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
+
+#Pure C for other kernels
+SAMAXKERNEL  = ../arm/amax.c
+DAMAXKERNEL  = ../arm/amax.c
+CAMAXKERNEL  = ../arm/zamax.c
+ZAMAXKERNEL  = ../arm/zamax.c
+
+SAMINKERNEL  = ../arm/amin.c
+DAMINKERNEL  = ../arm/amin.c
+CAMINKERNEL  = ../arm/zamin.c
+ZAMINKERNEL  = ../arm/zamin.c
+
+SMAXKERNEL   = ../arm/max.c
+DMAXKERNEL   = ../arm/max.c
+
+SMINKERNEL   = ../arm/min.c
+DMINKERNEL   = ../arm/min.c
+
+ISAMAXKERNEL = ../arm/iamax.c
+IDAMAXKERNEL = ../arm/iamax.c
+ICAMAXKERNEL = ../arm/izamax.c
+IZAMAXKERNEL = ../arm/izamax.c
+
+ISAMINKERNEL = ../arm/iamin.c
+IDAMINKERNEL = ../arm/iamin.c
+ICAMINKERNEL = ../arm/izamin.c
+IZAMINKERNEL = ../arm/izamin.c
+
+ISMAXKERNEL  = ../arm/imax.c
+IDMAXKERNEL  = ../arm/imax.c
+
+ISMINKERNEL  = ../arm/imin.c
+IDMINKERNEL  = ../arm/imin.c
+
+SASUMKERNEL  = ../arm/asum.c
+DASUMKERNEL  = ../arm/asum.c
+CASUMKERNEL  = ../arm/zasum.c
+ZASUMKERNEL  = ../arm/zasum.c
+
+SSUMKERNEL   = ../arm/sum.c
+DSUMKERNEL   = ../arm/sum.c
+CSUMKERNEL   = ../arm/zsum.c
+ZSUMKERNEL   = ../arm/zsum.c
+
+
+SAXPYKERNEL  = ../arm/axpy.c
+DAXPYKERNEL  = ../arm/axpy.c
+CAXPYKERNEL  = ../arm/zaxpy.c
+ZAXPYKERNEL  = ../arm/zaxpy.c
+
+SCOPYKERNEL  = ../arm/copy.c
+DCOPYKERNEL  = ../arm/copy.c
+CCOPYKERNEL  = ../arm/zcopy.c
+ZCOPYKERNEL  = ../arm/zcopy.c
+
+SDOTKERNEL   = ../generic/dot.c
+DDOTKERNEL   = ../arm/dot.c
+CDOTKERNEL   = ../arm/zdot.c
+ZDOTKERNEL   = ../arm/zdot.c
+
+SNRM2KERNEL  = ../arm/nrm2.c
+DNRM2KERNEL  = ../arm/nrm2.c
+CNRM2KERNEL  = ../arm/znrm2.c
+ZNRM2KERNEL  = ../arm/znrm2.c
+
+SROTKERNEL   = ../arm/rot.c
+DROTKERNEL   = ../arm/rot.c
+CROTKERNEL   = ../arm/zrot.c
+ZROTKERNEL   = ../arm/zrot.c
+
+SSCALKERNEL  = ../arm/scal.c
+DSCALKERNEL  = ../arm/scal.c
+CSCALKERNEL  = ../arm/zscal.c
+ZSCALKERNEL  = ../arm/zscal.c
+
+SSWAPKERNEL  = ../arm/swap.c
+DSWAPKERNEL  = ../arm/swap.c
+CSWAPKERNEL  = ../arm/zswap.c
+ZSWAPKERNEL  = ../arm/zswap.c
+
+SGEMVNKERNEL = ../arm/gemv_n.c
+DGEMVNKERNEL = ../arm/gemv_n.c
+CGEMVNKERNEL = ../arm/zgemv_n.c
+ZGEMVNKERNEL = ../arm/zgemv_n.c
+
+SGEMVTKERNEL = ../arm/gemv_t.c
+DGEMVTKERNEL = ../arm/gemv_t.c
+CGEMVTKERNEL = ../arm/zgemv_t.c
+ZGEMVTKERNEL = ../arm/zgemv_t.c
+
+SSYMV_U_KERNEL =  ../generic/symv_k.c
+SSYMV_L_KERNEL =  ../generic/symv_k.c
+DSYMV_U_KERNEL =  ../generic/symv_k.c
+DSYMV_L_KERNEL =  ../generic/symv_k.c
+QSYMV_U_KERNEL =  ../generic/symv_k.c
+QSYMV_L_KERNEL =  ../generic/symv_k.c
+CSYMV_U_KERNEL =  ../generic/zsymv_k.c
+CSYMV_L_KERNEL =  ../generic/zsymv_k.c
+ZSYMV_U_KERNEL =  ../generic/zsymv_k.c
+ZSYMV_L_KERNEL =  ../generic/zsymv_k.c
+XSYMV_U_KERNEL =  ../generic/zsymv_k.c
+XSYMV_L_KERNEL =  ../generic/zsymv_k.c
+
+ZHEMV_U_KERNEL =  ../generic/zhemv_k.c
+ZHEMV_L_KERNEL =  ../generic/zhemv_k.c
+
+LSAME_KERNEL   = ../generic/lsame.c
+SCABS_KERNEL   = ../generic/cabs.c
+DCABS_KERNEL   = ../generic/cabs.c
+QCABS_KERNEL   = ../generic/cabs.c
+
+#Dump kernel
+CGEMM3MKERNEL  = ../generic/zgemm3mkernel_dump.c
+ZGEMM3MKERNEL  = ../generic/zgemm3mkernel_dump.c

kernel/loongarch64/Makefile

@@ -0,0 +1 @@
+clean ::

kernel/loongarch64/amax.S

@@ -0,0 +1,230 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+
+#define I      $r17
+#define TEMP   $r18
+
+#define a1 $f10
+#define a2 $f11
+#define a3 $f12
+#define a4 $f13
+#define a5 $f14
+#define a6 $f15
+#define a7 $f16
+#define a8 $f17
+
+#define t1 $f0
+#define t2 $f1
+#define t3 $f2
+#define t4 $f3
+
+#define s1 $f22
+#define s2 $f8
+#define s3 $f23
+#define s4 $f9
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+
+   LD a1,  X,   0 * SIZE
+   addi.d  N, N, -1
+
+   add.d   X, X, INCX
+   FABS    s1, a1
+
+   FABS   s2, a1
+   bge $r0,    N, .L999
+
+   FABS    s3, a1
+   srai.d  I, N, 3
+
+   FABS   s4, a1
+   bge $r0,    I, .L15
+
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   addi.d  I, I, -1
+
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+
+   FABS    t3, a3
+   LD a2,  X,   0 * SIZE
+   FABS    t4, a4
+   add.d   X, X, INCX
+
+   CMPLT   $fcc0, s1, t1
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t2
+   add.d   X, X, INCX
+
+   CMPLT   $fcc2, s3, t3
+   LD a4,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, t4
+   add.d   X, X, INCX
+
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+
+   FABS    t1, a5
+   LD a5,  X,   0 * SIZE
+   FABS    t2, a6
+   add.d   X, X, INCX
+
+   FABS    t3, a7
+   LD a6,  X,   0 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+
+   CMPLT   $fcc0, s1, t1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t2
+   add.d   X, X, INCX
+
+   CMPLT   $fcc2, s3, t3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, t4
+   add.d   X, X, INCX
+
+   CMOVT  s1,  s1,  t1,  $fcc0
+   addi.d  I, I, -1
+
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+
+   CMOVT  s4,  s4,  t4,  $fcc3
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMPLT   $fcc2, s3, t3
+   CMPLT   $fcc3, s4, t4
+
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+
+   FABS    t1, a5
+   FABS    t2, a6
+   FABS    t3, a7
+   FABS    t4, a8
+
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMPLT   $fcc2, s3, t3
+   CMPLT   $fcc3, s4, t4
+
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+
+   FABS    t1, a1
+
+   CMPLT   $fcc0, s1, t1
+
+   CMOVT  s1,  s1,  t1,  $fcc0
+
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s1, s2
+   CMPLT   $fcc1, s3, s4
+
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+
+   CMPLT   $fcc0, s1, s3
+   CMOVT  s1,  s1,  s3,  $fcc0
+   .align 3
+
+.L999:
+  move $r4, $r17
+  fmov.d $f0, $f22
+  jirl    $r0, $r1, 0x0
+
+  EPILOGUE

kernel/loongarch64/amin.S

@@ -0,0 +1,186 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+
+   PROLOGUE
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD a1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   add.d   X, X, INCX
+   FABS    s1, a1
+   FABS    s2, a1
+   bge     $r0,    N, .L999
+   FABS    s3, a1
+   srai.d  I, N, 3
+   FABS   s4, a1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+   FABS    t3, a3
+   LD a2,  X,   0 * SIZE
+   FABS    t4, a4
+   add.d   X, X, INCX
+   CMPLT   $fcc0, t1, s1
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, t2, s2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, t3, s3
+   LD a4,  X,   0 * SIZE
+   CMPLT   $fcc3, t4, s4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   FABS    t1, a5
+   LD a5,  X,   0 * SIZE
+   FABS    t2, a6
+   add.d   X, X, INCX
+   FABS    t3, a7
+   LD a6,  X,   0 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+   CMPLT   $fcc0, t1, s1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, t2, s2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, t3, s3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, t4, s4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   addi.d  I, I, -1
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   blt $r0,    I, .L12
+   .align 3
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   CMPLT   $fcc0, t1, s1
+   CMPLT   $fcc1, t2, s2
+   CMPLT   $fcc2, t3, s3
+   CMPLT   $fcc3, t4, s4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   FABS    t1, a5
+   FABS    t2, a6
+   FABS    t3, a7
+   FABS    t4, a8
+   CMPLT   $fcc0, t1, s1
+   CMPLT   $fcc1, t2, s2
+   CMPLT   $fcc2, t3, s3
+   CMPLT   $fcc3, t4, s4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   .align 3
+.L15:
+   andi    I,  N, 7
+NOP
+   bge $r0,    I, .L998
+   .align  3
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   FABS    t1, a1
+   CMPLT   $fcc0, t1, s1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+.L998:
+   CMPLT   $fcc0, s2, s1
+   CMPLT   $fcc1, s4, s3
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+   CMPLT   $fcc0, s3, s1
+   CMOVT  s1,  s1,  s3,  $fcc0
+   .align 3
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   EPILOGUE

kernel/loongarch64/asum.S

@@ -0,0 +1,232 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1 $f23
+#define a2 $f9
+#define a3 $f10
+#define a4 $f11
+#define a5 $f12
+#define a6 $f13
+#define a7 $f14
+#define a8 $f15
+#define t1 $f16
+#define t2 $f17
+#define t3 $f0
+#define t4 $f1
+#define s1 $f22
+#define s2 $f8
+   PROLOGUE
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+   MTC  s1, $r0
+   MTC  s2, $r0
+   slli.d  INCX, INCX, BASE_SHIFT
+   li TEMP, SIZE
+   bge $r0,    N, .L999
+   srai.d  I, N, 3
+   bne INCX, TEMP, .L20
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   LD a5,  X,   4 * SIZE
+   FABS    t1, a1
+   LD a6,  X,   5 * SIZE
+   FABS    t2, a2
+   LD a7,  X,   6 * SIZE
+   FABS    t3, a3
+   FABS    t4, a4
+   addi.d  I, I, -1
+   LD a8,  X,   7 * SIZE
+   bge $r0,    I, .L13
+   .align 3
+.L12:
+   ADD s1, s1, t1
+   LD a1,  X,   8 * SIZE
+   FABS    t1, a5
+   addi.d  I, I, -1
+   ADD s2, s2, t2
+   LD a2,  X,   9 * SIZE
+   FABS    t2, a6
+   NOP
+   ADD s1, s1, t3
+   LD a3,  X,  10 * SIZE
+   FABS    t3, a7
+   NOP
+   ADD s2, s2, t4
+   LD a4,  X,  11 * SIZE
+   FABS    t4, a8
+   addi.d  X, X, 8 * SIZE
+   ADD s1, s1, t1
+   LD a5,  X,   4 * SIZE
+   FABS    t1, a1
+   NOP
+   ADD s2, s2, t2
+   LD a6,  X,   5 * SIZE
+   FABS    t2, a2
+   NOP
+   ADD s1, s1, t3
+   LD a7,  X,   6 * SIZE
+   FABS    t3, a3
+   NOP
+   ADD s2, s2, t4
+   LD a8,  X,   7 * SIZE
+   FABS   t4, a4
+   blt $r0,    I, .L12
+   .align 3
+.L13:
+   ADD s1, s1, t1
+   addi.d  X, X, 8 * SIZE
+   FABS    t1, a5
+   NOP
+   ADD s2, s2, t2
+   FABS    t2, a6
+   ADD s1, s1, t3
+   FABS    t3, a7
+   ADD s2, s2, t4
+   FABS    t4, a8
+   ADD s1, s1, t1
+   ADD s2, s2, t2
+   ADD s1, s1, t3
+   ADD s2, s2, t4
+   .align 3
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   FABS    t1, a1
+   ADD s1, s1, t1
+   addi.d X, X, SIZE
+   blt $r0,    I, .L16
+   b   .L999
+   .align 3
+.L20:
+   bge $r0,    I, .L25
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   LD a7,  X,   0 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+   FABS    t3, a3
+   LD a8,  X,   0 * SIZE
+   FABS    t4, a4
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L24
+   .align 3
+.L23:
+   ADD s1, s1, t1
+   LD a1,  X,   0 * SIZE
+   FABS    t1, a5
+   add.d   X, X, INCX
+   ADD s2, s2, t2
+   LD a2,  X,   0 * SIZE
+   FABS    t2, a6
+   add.d   X, X, INCX
+   ADD s1, s1, t3
+   LD a3,  X,   0 * SIZE
+   FABS    t3, a7
+   add.d   X, X, INCX
+   ADD s2, s2, t4
+   LD a4,  X,   0 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+   ADD s1, s1, t1
+   LD a5,  X,   0 * SIZE
+   FABS    t1, a1
+   add.d   X, X, INCX
+   ADD s2, s2, t2
+   LD a6,  X,   0 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+   ADD s1, s1, t3
+   LD a7,  X,   0 * SIZE
+   FABS    t3, a3
+   add.d   X, X, INCX
+   ADD s2, s2, t4
+   LD a8,  X,   0 * SIZE
+   FABS    t4, a4
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L23
+   .align 3
+.L24:
+   ADD s1, s1, t1
+   FABS    t1, a5
+   ADD s2, s2, t2
+   FABS    t2, a6
+   ADD s1, s1, t3
+   FABS    t3, a7
+   ADD s2, s2, t4
+   FABS    t4, a8
+   ADD s1, s1, t1
+   ADD s2, s2, t2
+   ADD s1, s1, t3
+   ADD s2, s2, t4
+   .align 3
+.L25:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L26:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   FABS    t1, a1
+   add.d   X, X, INCX
+   ADD    s1, s1, t1
+   blt $r0,    I, .L26
+   .align 3
+.L999:
+   ADD s1, s1, s2
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   EPILOGUE

kernel/loongarch64/cnrm2.S

@@ -0,0 +1,159 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f12
+#define a2     $f13
+#define a3     $f14
+#define a4     $f15
+#define a5     $f16
+#define a6     $f17
+#define a7     $f0
+#define a8     $f1
+#define s1     $f22
+#define s2     $f8
+#define t1     $f23
+#define t2     $f9
+#define t3     $f10
+#define t4     $f11
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   movgr2fr.d  s1,  $r0
+   li  TEMP, 2 * SIZE
+   fmov.d s2, s1
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    INCX, .L999
+   srai.d  I, N, 2
+   bge $r0,    I, .L25
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   fcvt.d.s    t1, a1
+   LD a7,  X,   0 * SIZE
+   fcvt.d.s    t2, a2
+   LD a8,  X,   1 * SIZE
+   fcvt.d.s    t3, a3
+   addi.d  I, I, -1
+   fcvt.d.s    t4, a4
+   add.d  X, X, INCX
+   bge $r0,    I, .L24
+   .align 3
+
+.L23:
+   fmadd.d  s1, t1, t1, s1
+   LD a1,  X,   0 * SIZE
+   fcvt.d.s    t1, a5
+   fmadd.d  s2, t2, t2, s2
+   LD a2,  X,   1 * SIZE
+   fcvt.d.s    t2, a6
+   add.d   X, X, INCX
+   fmadd.d  s1, t3, t3, s1
+   LD a3,  X,   0 * SIZE
+   fcvt.d.s    t3, a7
+   fmadd.d  s2, t4, t4, s2
+   LD a4,  X,   1 * SIZE
+   fcvt.d.s    t4, a8
+   add.d   X, X, INCX
+   fmadd.d  s1, t1, t1, s1
+   LD a5,  X,   0 * SIZE
+   fcvt.d.s    t1, a1
+   addi.d  I, I, -1
+   fmadd.d  s2, t2, t2, s2
+   LD a6,  X,   1 * SIZE
+   fcvt.d.s    t2, a2
+   add.d   X, X, INCX
+   fmadd.d  s1, t3, t3, s1
+   LD a7,  X,   0 * SIZE
+   fcvt.d.s    t3, a3
+   LD a8,  X,   1 * SIZE
+   fmadd.d  s2, t4, t4, s2
+   add.d   X, X, INCX
+   fcvt.d.s   t4, a4
+   blt $r0,    I, .L23
+   .align 3
+
+.L24:
+   fmadd.d  s1, t1, t1, s1
+   fcvt.d.s    t1, a5
+   fmadd.d  s2, t2, t2, s2
+   fcvt.d.s    t2, a6
+   fmadd.d  s1, t3, t3, s1
+   fcvt.d.s    t3, a7
+   fmadd.d  s2, t4, t4, s2
+   fcvt.d.s    t4, a8
+   fmadd.d  s1, t1, t1, s1
+   fmadd.d  s2, t2, t2, s2
+   fmadd.d  s1, t3, t3, s1
+   fmadd.d  s2, t4, t4, s2
+   .align 3
+
+.L25:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   .align  3
+
+.L26:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   addi.d  I, I, -1
+   fcvt.d.s    t1, a1
+   fcvt.d.s    t2, a2
+   fmadd.d  s1, t1, t1, s1
+   add.d   X, X, INCX
+   fmadd.d  s2, t2, t2, s2
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+   fadd.d  s1, s1, s2
+   fsqrt.d s1, s1
+   move $r4, $r17
+   fcvt.s.d    $f0, s1
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/copy.S

@@ -0,0 +1,225 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define Y      $r7
+#define INCY   $r8
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f22
+#define a2     $f8
+#define a3     $f23
+#define a4     $f9
+#define a5     $f10
+#define a6     $f11
+#define a7     $f12
+#define a8     $f13
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+   LDINT   INCY,  0(INCY)
+#endif
+
+   li  TEMP, SIZE
+   NOP
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    N, .L999
+   slli.d INCY, INCY, BASE_SHIFT
+   bne INCX, TEMP, .L20
+   srai.d I, N, 3
+   bne INCY, TEMP, .L20
+   addi.d I, I, -1
+   blt     I, $r0,     .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   LD a5,  X,   4 * SIZE
+   LD a6,  X,   5 * SIZE
+   LD a7,  X,   6 * SIZE
+   LD a8,  X,   7 * SIZE
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   ST a1,  Y,   0 * SIZE
+   LD a1,  X,   8 * SIZE
+   ST a2,  Y,   1 * SIZE
+   LD a2,  X,   9 * SIZE
+   ST a3,  Y,   2 * SIZE
+   LD a3,  X,  10 * SIZE
+   ST a4,  Y,   3 * SIZE
+   LD a4,  X,  11 * SIZE
+   ST a5,  Y,   4 * SIZE
+   LD a5,  X,  12 * SIZE
+   ST a6,  Y,   5 * SIZE
+   LD a6,  X,  13 * SIZE
+   ST a7,  Y,   6 * SIZE
+   LD a7,  X,  14 * SIZE
+   ST a8,  Y,   7 * SIZE
+   LD a8,  X,  15 * SIZE
+   addi.d  I, I, -1
+   addi.d  X, X, 8 * SIZE
+   addi.d Y, Y, 8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   ST a1,  Y,   0 * SIZE
+   ST a2,  Y,   1 * SIZE
+   ST a3,  Y,   2 * SIZE
+   ST a4,  Y,   3 * SIZE
+   ST a5,  Y,   4 * SIZE
+   ST a6,  Y,   5 * SIZE
+   ST a7,  Y,   6 * SIZE
+   ST a8,  Y,   7 * SIZE
+   addi.d  X, X, 8 * SIZE
+   addi.d  Y, Y, 8 * SIZE
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  X, X, SIZE
+   addi.d  I, I, -1
+   addi.d  Y, Y, SIZE
+   ST a1,  Y,  -1 * SIZE
+   blt $r0,    I, .L16
+   b   .L999
+   .align 3
+
+.L20:
+   srai.d  I, N, 3
+   addi.d I, I, -1
+   blt I,  $r0, .L25
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   add.d   X, X, INCX
+   bge $r0,    I, .L23
+   .align 3
+
+.L22:
+   ST a1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a2,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a3,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a4,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a5,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a6,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a7,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a8,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a8,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L22
+   .align 3
+
+.L23:
+   ST a1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a2,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a3,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a4,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a5,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a6,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a7,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a8,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   .align 3
+
+.L25:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+
+.L26:
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   addi.d  I, I, -1
+   ST a1,  Y,   0 * SIZE
+   add.d  Y, Y, INCY
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/dnrm2.S

@@ -0,0 +1,314 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define XX     $r7
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+#define ALPHA  $f4
+#define max    $f5
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+   move    XX, X
+   NOP
+   LD a1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   add.d   X, X, INCX
+   FABS    s1, a1
+   FABS   s2, a1
+   bge $r0,    N, .L999
+   FABS    s3, a1
+   srai.d  I, N, 3
+   FABS   s4, a1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+   FABS    t3, a3
+   LD a2,  X,   0 * SIZE
+   FABS    t4, a4
+   add.d   X, X, INCX
+   CMPLT   $fcc0, s1, t1
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, s3, t3
+   LD a4,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, t4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   FABS    t1, a5
+   LD a5,  X,   0 * SIZE
+   FABS    t2, a6
+   add.d   X, X, INCX
+   FABS    t3, a7
+   LD a6,  X,   0 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+   CMPLT   $fcc0, s1, t1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, s3, t3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, t4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   addi.d  I, I, -1
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMPLT   $fcc2, s3, t3
+   CMPLT   $fcc3, s4, t4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   FABS    t1, a5
+   FABS    t2, a6
+   FABS    t3, a7
+   FABS    t4, a8
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMPLT   $fcc2, s3, t3
+   CMPLT   $fcc3, s4, t4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L100
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   FABS    t1, a1
+   CMPLT   $fcc0, s1, t1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+
+.L100:
+   CMPLT   $fcc0, s1, s2
+   CMPLT   $fcc1, s3, s4
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+   CMPLT   $fcc0, s1, s3
+   CMOVT  s1,  s1,  s3,  $fcc0
+   addi.d  N, N, 1
+   lu12i.w TEMP, 0x3f800
+   movgr2fr.d      a1,   $r0
+   movgr2fr.w  ALPHA, TEMP
+   CMPEQ   $fcc0, s1, a1
+   fcvt.d.s   ALPHA, ALPHA
+   bcnez   $fcc0, .L999
+   fdiv.d  ALPHA, ALPHA, s1
+   MOV max, s1
+   MOV s1, a1
+   MOV s2, a1
+   MOV s3, a1
+   MOV s4, a1
+   srai.d  I, N, 3
+   bge $r0,    I, .L105
+   LD a1,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD a2,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD a3,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD a4,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD a5,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD a6,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD a7,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD a8,  XX,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  XX, XX, INCX
+   bge $r0,    I, .L104
+   .align 3
+
+.L103:
+   MUL t1, ALPHA, a1
+   LD a1,  XX,   0 * SIZE
+   MUL t2, ALPHA, a2
+   add.d   XX, XX, INCX
+   MUL t3, ALPHA, a3
+   LD a2,  XX,   0 * SIZE
+   MUL t4, ALPHA, a4
+   add.d   XX, XX, INCX
+   MADD  s1, t1, t1, s1
+   LD a3,  XX,   0 * SIZE
+   MADD  s2, t2, t2, s2
+   add.d   XX, XX, INCX
+   MADD  s3, t3, t3, s3
+   LD a4,  XX,   0 * SIZE
+   MADD  s4, t4, t4, s4
+   add.d   XX, XX, INCX
+   MUL t1, ALPHA, a5
+   LD a5,  XX,   0 * SIZE
+   MUL t2, ALPHA, a6
+   add.d   XX, XX, INCX
+   MUL t3, ALPHA, a7
+   LD a6,  XX,   0 * SIZE
+   MUL t4, ALPHA, a8
+   add.d   XX, XX, INCX
+   MADD  s1, t1, t1, s1
+   LD a7,  XX,   0 * SIZE
+   MADD  s2, t2, t2, s2
+   add.d   XX, XX, INCX
+   MADD  s3, t3, t3, s3
+   LD a8,  XX,   0 * SIZE
+   MADD  s4, t4, t4, s4
+   addi.d  I, I, -1
+   add.d  XX, XX, INCX
+   blt $r0,    I, .L103
+   .align 3
+
+.L104:
+   MUL t1, ALPHA, a1
+   MUL t2, ALPHA, a2
+   MUL t3, ALPHA, a3
+   MUL t4, ALPHA, a4
+   MADD  s1, t1, t1, s1
+   MADD  s2, t2, t2, s2
+   MADD  s3, t3, t3, s3
+   MADD  s4, t4, t4, s4
+   MUL t1, ALPHA, a5
+   MUL t2, ALPHA, a6
+   MUL t3, ALPHA, a7
+   MUL t4, ALPHA, a8
+   MADD  s1, t1, t1, s1
+   MADD  s2, t2, t2, s2
+   MADD  s3, t3, t3, s3
+   MADD  s4, t4, t4, s4
+   .align 3
+
+.L105:
+   andi    I,  N, 7
+   bge $r0,    I, .L998
+   .align  3
+
+.L106:
+   LD a1,  XX,   0 * SIZE
+   addi.d  I, I, -1
+   MUL t1, ALPHA, a1
+   add.d   XX, XX, INCX
+   MADD  s1, t1, t1, s1
+   blt $r0,    I, .L106
+   .align 3
+
+.L998:
+   ADD s1, s1, s2
+   ADD s3, s3, s4
+   ADD s1, s1, s3
+   fsqrt.d s1, s1
+   move $r4, $r17
+   MUL $f0, max, s1
+   jirl    $r0, $r1, 0x0
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/dot.S

@@ -0,0 +1,391 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define Y      $r7
+#define INCY   $r8
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f23
+#define a2     $f9
+#define a3     $f10
+#define a4     $f11
+#define b1     $f12
+#define b2     $f13
+#define b3     $f14
+#define b4     $f15
+#define s1     $f22
+#define s2     $f8
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+   LDINT   INCY,  0(INCY)
+#endif
+
+   MTC  s1, $r0
+   MTC  s2, $r0
+   slli.d  INCX, INCX, BASE_SHIFT
+   li  TEMP, SIZE
+   slli.d INCY, INCY, BASE_SHIFT
+   bge $r0,    N, .L999
+   srai.d I, N, 3
+   bne INCX, TEMP, .L20
+   bne INCY, TEMP, .L20
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD b1,  Y,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD b2,  Y,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD b3,  Y,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   addi.d  I, I, -1
+   LD b4,  Y,   3 * SIZE
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   LD a1,  X,   4 * SIZE
+   LD b1,  Y,   4 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a2, a2
+   fcvt.d.s  b2, b2
+   fmadd.d  s2, b2, a2, s2
+#else
+   MADD  s2, b2, a2, s2
+#endif
+   LD a2,  X,   5 * SIZE
+   LD b2,  Y,   5 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a3, a3
+   fcvt.d.s  b3, b3
+   fmadd.d  s1, b3, a3, s1
+#else
+   MADD  s1, b3, a3, s1
+#endif
+   LD a3,  X,   6 * SIZE
+   LD b3,  Y,   6 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a4, a4
+   fcvt.d.s  b4, b4
+   fmadd.d  s2, b4, a4, s2
+#else
+   MADD  s2, b4, a4, s2
+#endif
+   LD a4,  X,   7 * SIZE
+   LD b4,  Y,   7 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   LD a1,  X,   8 * SIZE
+   LD b1,  Y,   8 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a2, a2
+   fcvt.d.s  b2, b2
+   fmadd.d  s2, b2, a2, s2
+#else
+   MADD  s2, b2, a2, s2
+#endif
+   LD a2,  X,   9 * SIZE
+   LD b2,  Y,   9 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a3, a3
+   fcvt.d.s  b3, b3
+   fmadd.d  s1, b3, a3, s1
+#else
+   MADD  s1, b3, a3, s1
+#endif
+   LD a3,  X,  10 * SIZE
+   LD b3,  Y,  10 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a4, a4
+   fcvt.d.s  b4, b4
+   fmadd.d  s2, b4, a4, s2
+#else
+   MADD  s2, b4, a4, s2
+#endif
+   LD a4,  X,  11 * SIZE
+   LD b4,  Y,  11 * SIZE
+   addi.d  I, I, -1
+   addi.d  X, X, 8 * SIZE
+addi.d Y, Y, 8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+.L13:
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   LD a1,  X,   4 * SIZE
+   LD b1,  Y,   4 * SIZE
+#ifdef DSDOT
+   fcvt.d.s a2, a2
+   fcvt.d.s b2, b2
+   fmadd.d  s2, b2, a2, s2
+#else
+   MADD  s2, b2, a2, s2
+#endif
+   LD a2,  X,   5 * SIZE
+   LD b2,  Y,   5 * SIZE
+#ifdef DSDOT
+   fcvt.d.s a3, a3
+   fcvt.d.s b3, b3
+   fmadd.d  s1, b3, a3, s1
+#else
+   MADD  s1, b3, a3, s1
+#endif
+   LD a3,  X,   6 * SIZE
+   LD b3,  Y,   6 * SIZE
+#ifdef DSDOT
+   fcvt.d.s a4, a4
+   fcvt.d.s b4, b4
+   fmadd.d  s2, b4, a4, s2
+#else
+   MADD  s2, b4, a4, s2
+#endif
+   LD a4,  X,   7 * SIZE
+   LD b4,  Y,   7 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   addi.d  X, X, 8 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a2, a2
+   fcvt.d.s  b2, b2
+   fmadd.d  s2, b2, a2, s2
+#else
+   MADD  s2, b2, a2, s2
+#endif
+   addi.d  Y, Y, 8 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a3, a3
+   fcvt.d.s  b3, b3
+   fmadd.d  s1, b3, a3, s1
+#else
+   MADD  s1, b3, a3, s1
+#endif
+#ifdef DSDOT
+   fcvt.d.s  a4, a4
+   fcvt.d.s  b4, b4
+   fmadd.d  s2, b4, a4, s2
+#else
+   MADD  s2, b4, a4, s2
+#endif
+   .align 3
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD b1,  Y,   0 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   addi.d  I, I, -1
+   addi.d  X, X, SIZE
+   addi.d  Y, Y, SIZE
+   blt $r0,    I, .L16
+   b   .L999
+   .align 3
+
+.L20:
+#ifdef F_INTERFACE
+   bgez    INCX, .L21
+   addi.d  TEMP, N, -1
+   mult    TEMP, INCX
+   mflo    TEMP
+   dsub    X, X, TEMP
+   .align 3
+
+.L21:
+   bgez    INCY, .L22
+   addi.d  TEMP, N, -1
+   mult    TEMP, INCY
+   mflo    TEMP
+   dsub    Y, Y, TEMP
+   .align 3
+
+.L22:
+#endif
+   bge $r0,    I, .L25
+   .align 3
+
+.L23:
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s2, b1, a1, s2
+#else
+   MADD  s2, b1, a1, s2
+#endif
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s2, b1, a1, s2
+#else
+   MADD  s2, b1, a1, s2
+#endif
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s2, b1, a1, s2
+#else
+   MADD  s2, b1, a1, s2
+#endif
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   addi.d  I, I, -1
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s2, b1, a1, s2
+#else
+   MADD  s2, b1, a1, s2
+#endif
+   blt $r0,    I, .L23
+   .align 3
+
+.L25:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+
+.L26:
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   addi.d  I, I, -1
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+#ifdef DSDOT
+   fadd.d $f0, s1, s2
+#else
+   ADD    $f0, s1, s2
+#endif
+   move $r4, $r17
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/gemv_n.S

@@ -0,0 +1,531 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+/* Unused param dummy1 */
+#define M      $r4
+#define N      $r5
+#define A      $r7
+#define LDA    $r8
+#define X      $r9
+#define INCX   $r10
+#define Y      $r11
+#define INCY   $r6
+#define BUFFER $r16
+#define YORIG  $r18
+#define XX     $r12
+#define YY     $r13
+#define I      $r14
+#define J      $r15
+#define AO1    $r23
+#define AO2    $r24
+#define ALPHA  $f0
+#define a1     $f22
+#define a2     $f8
+#define a3     $f23
+#define a4     $f9
+#define a5     $f10
+#define a6     $f11
+#define a7     $f12
+#define a8     $f13
+#define x1     $f14
+#define x2     $f15
+#define y1     $f16
+#define y2     $f17
+#define y3     $f3
+#define y4     $f1
+#define y5     $f2
+#define y6     $f4
+#define y7     $f5
+#define y8     $f6
+#define t1     $f7
+#define t2     $f18
+#define t3     $f19
+#define t4     $f20
+
+   PROLOGUE
+
+   LDARG  INCY,    $sp,  0
+   LDARG  BUFFER,  $sp,  8
+#ifdef __64BIT__
+   addi.d  $sp, $sp, -16
+#else
+   addi.d  $sp, $sp, -48
+#endif
+   SDARG  $r23,  $sp,    0
+   SDARG  $r24,  $sp,    8
+   slli.d     LDA,     LDA,  BASE_SHIFT
+#ifndef __64BIT__
+   fst.d  $f18,  $sp,  16
+   fst.d  $f19,  $sp,  24
+   fst.d  $f20,  $sp,  32
+#endif
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    M, .L999
+   slli.d INCY, INCY, BASE_SHIFT
+   bge $r0,    N, .L999
+   li  I, SIZE
+   move   YORIG, Y
+   beq INCY, I, .L10
+   srai.d  I,  M, 2
+   move    YORIG, BUFFER
+   move    XX, Y
+   move   YY, BUFFER
+   bge $r0,    I, .L05
+   .align 3
+
+.L02:
+   LD a1,  XX,  0 * SIZE
+   add.d   XX, XX, INCY
+   LD a2,  XX,  0 * SIZE
+   add.d   XX, XX, INCY
+   LD a3,  XX,  0 * SIZE
+   add.d   XX, XX, INCY
+   LD a4,  XX,  0 * SIZE
+   add.d   XX, XX, INCY
+   ST a1,  YY,  0 * SIZE
+   ST a2,  YY,  1 * SIZE
+   ST a3,  YY,  2 * SIZE
+   ST a4,  YY,  3 * SIZE
+   addi.d  I, I, -1
+   addi.d YY, YY, 4 * SIZE
+   blt $r0,    I, .L02
+   .align 3
+
+.L05:
+   andi    I,  M, 3
+   bge $r0,    I, .L10
+   .align 3
+
+.L06:
+   LD a1,  XX,  0 * SIZE
+   add.d   XX, XX, INCY
+   ST a1,  YY,  0 * SIZE
+   addi.d  I, I, -1
+   addi.d YY, YY, 1 * SIZE
+   blt $r0,    I, .L06
+   .align 3
+
+.L10:
+   srai.d  J,  N, 1
+   bge $r0,    J, .L20
+   .align 3
+
+.L11:
+   LD x1,  X,  0 * SIZE
+   add.d   X, X, INCX
+   LD x2,  X,  0 * SIZE
+   add.d   X, X, INCX
+   move    AO1, A
+   add.d   AO2, A,      LDA
+   add.d   A,   AO2,    LDA
+   move    YY, YORIG
+   MUL x1, ALPHA, x1
+   srai.d  I,  M, 3
+   MUL    x2, ALPHA, x2
+   bge $r0,    I, .L15
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   LD y4,  YY,  3 * SIZE
+   LD a5,  AO2,  0 * SIZE
+   LD y5,  YY,  4 * SIZE
+   LD a6,  AO2,  1 * SIZE
+   LD y6,  YY,  5 * SIZE
+   LD a7,  AO2,  2 * SIZE
+   LD y7,  YY,  6 * SIZE
+   LD a8,  AO2,  3 * SIZE
+   addi.d  I, I, -1
+   LD y8,  YY,  7 * SIZE
+   bge $r0,    I, .L13
+   .align  3
+.L12:
+   MADD  t1, a1, x1, y1
+   LD a1,  AO1,   4 * SIZE
+   MADD  t2, a2, x1, y2
+   LD a2,  AO1,   5 * SIZE
+   LD y1,  YY,   8 * SIZE
+   LD y2,  YY,   9 * SIZE
+   MADD  t3, a3, x1, y3
+   LD a3,  AO1,   6 * SIZE
+   MADD  t4, a4, x1, y4
+   LD a4,  AO1,   7 * SIZE
+   LD y3,  YY,  10 * SIZE
+   LD y4,  YY,  11 * SIZE
+   MADD  t1, a5, x2, t1
+   LD a5,  AO2,   4 * SIZE
+   MADD  t2, a6, x2, t2
+   LD a6,  AO2,   5 * SIZE
+   MADD  t3, a7, x2, t3
+   LD a7,  AO2,   6 * SIZE
+   MADD  t4, a8, x2, t4
+   LD a8,  AO2,   7 * SIZE
+   ST t1,  YY,   0 * SIZE
+   ST t2,  YY,   1 * SIZE
+   ST t3,  YY,   2 * SIZE
+   ST t4,  YY,   3 * SIZE
+   MADD  t1, a1, x1, y5
+   LD a1,  AO1,   8 * SIZE
+   MADD  t2, a2, x1, y6
+   LD a2,  AO1,   9 * SIZE
+   LD y5,  YY,  12 * SIZE
+   LD y6,  YY,  13 * SIZE
+   MADD  t3, a3, x1, y7
+   LD a3,  AO1,  10 * SIZE
+   MADD  t4, a4, x1, y8
+   LD a4,  AO1,  11 * SIZE
+   LD y7,  YY,  14 * SIZE
+   LD y8,  YY,  15 * SIZE
+   MADD  t1, a5, x2, t1
+   LD a5,  AO2,   8 * SIZE
+   MADD  t2, a6, x2, t2
+   LD a6,  AO2,   9 * SIZE
+   MADD  t3, a7, x2, t3
+   LD a7,  AO2,  10 * SIZE
+   MADD  t4, a8, x2, t4
+   LD a8,  AO2,  11 * SIZE
+   ST t1,  YY,  4 * SIZE
+   ST t2,  YY,  5 * SIZE
+   ST t3,  YY,  6 * SIZE
+   ST t4,  YY,  7 * SIZE
+   addi.d  I, I, -1
+   addi.d  YY,  YY,   8 * SIZE
+   addi.d  AO1, AO1,  8 * SIZE
+   addi.d AO2, AO2,  8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   MADD  t1, a1, x1, y1
+   LD a1,  AO1,   4 * SIZE
+   MADD  t2, a2, x1, y2
+   LD a2,  AO1,   5 * SIZE
+   MADD  t3, a3, x1, y3
+   LD a3,  AO1,   6 * SIZE
+   MADD  t4, a4, x1, y4
+   LD a4,  AO1,   7 * SIZE
+   MADD  t1, a5, x2, t1
+   LD a5,  AO2,   4 * SIZE
+   MADD  t2, a6, x2, t2
+   LD a6,  AO2,   5 * SIZE
+   MADD  t3, a7, x2, t3
+   LD a7,  AO2,   6 * SIZE
+   MADD  t4, a8, x2, t4
+   LD a8,  AO2,   7 * SIZE
+   ST t1,  YY,   0 * SIZE
+   MADD  t1, a1, x1, y5
+   ST t2,  YY,   1 * SIZE
+   MADD  t2, a2, x1, y6
+   ST t3,  YY,   2 * SIZE
+   MADD  t3, a3, x1, y7
+   ST t4,  YY,   3 * SIZE
+   MADD  t4, a4, x1, y8
+   MADD  t1, a5, x2, t1
+   addi.d  AO1, AO1,  8 * SIZE
+   MADD  t2, a6, x2, t2
+   addi.d  AO2, AO2,  8 * SIZE
+   MADD  t3, a7, x2, t3
+   addi.d  YY,  YY,   8 * SIZE
+   MADD  t4, a8, x2, t4
+   ST t1,  YY,  -4 * SIZE
+   ST t2,  YY,  -3 * SIZE
+   ST t3,  YY,  -2 * SIZE
+   ST t4,  YY,  -1 * SIZE
+   .align 3
+
+.L15:
+   andi    I,  M, 4
+   bge $r0,    I, .L16
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   LD y4,  YY,  3 * SIZE
+   LD a5,  AO2,  0 * SIZE
+   MADD  y1, a1, x1, y1
+   LD a6,  AO2,  1 * SIZE
+   MADD  y2, a2, x1, y2
+   LD a7,  AO2,  2 * SIZE
+   MADD  y3, a3, x1, y3
+   LD a8,  AO2,  3 * SIZE
+   MADD  y4, a4, x1, y4
+   MADD  y1, a5, x2, y1
+   addi.d  YY,  YY,   4 * SIZE
+   MADD  y2, a6, x2, y2
+   addi.d  AO1, AO1,  4 * SIZE
+   MADD  y3, a7, x2, y3
+   addi.d  AO2, AO2,  4 * SIZE
+   MADD  y4, a8, x2, y4
+   ST y1,  YY,  -4 * SIZE
+   ST y2,  YY,  -3 * SIZE
+   ST y3,  YY,  -2 * SIZE
+   ST y4,  YY,  -1 * SIZE
+   .align 3
+
+.L16:
+   andi    I,  M, 2
+   bge $r0,    I, .L17
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a5,  AO2,  0 * SIZE
+   LD a6,  AO2,  1 * SIZE
+   MADD  y1, a1, x1, y1
+   MADD  y2, a2, x1, y2
+   addi.d  YY,  YY,   2 * SIZE
+   MADD  y1, a5, x2, y1
+   addi.d  AO1, AO1,  2 * SIZE
+   MADD  y2, a6, x2, y2
+   addi.d  AO2, AO2,  2 * SIZE
+   ST y1,  YY,  -2 * SIZE
+   ST y2,  YY,  -1 * SIZE
+   .align 3
+
+.L17:
+   andi    I,  M, 1
+   bge $r0,    I, .L19
+   LD y1,  YY,  0 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD a5,  AO2,  0 * SIZE
+   MADD  y1, a1, x1, y1
+   MADD  y1, a5, x2, y1
+   ST y1,  YY,  0 * SIZE
+   .align 3
+
+.L19:
+   addi.d  J, J, -1
+   blt $r0,    J, .L11
+   .align 3
+
+.L20:
+   andi    J,  N, 1
+   bge $r0,    J, .L900
+   .align 3
+
+.L21:
+   LD x1,  X,  0 * SIZE
+   add.d   X, X, INCX
+   move    YY, YORIG
+   move    AO1, A
+   srai.d  I,  M, 3
+   MUL    x1, ALPHA, x1
+   bge $r0,    I, .L25
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   LD y4,  YY,  3 * SIZE
+   LD y5,  YY,  4 * SIZE
+   LD y6,  YY,  5 * SIZE
+   LD y7,  YY,  6 * SIZE
+   addi.d  I, I, -1
+   LD y8,  YY,  7 * SIZE
+   bge $r0,    I, .L23
+   .align  3
+.L22:
+   MADD  t1, a1, x1, y1
+   LD a1,  AO1,   4 * SIZE
+   MADD  t2, a2, x1, y2
+   LD a2,  AO1,   5 * SIZE
+   LD y1,  YY,   8 * SIZE
+   LD y2,  YY,   9 * SIZE
+   MADD  t3, a3, x1, y3
+   LD a3,  AO1,   6 * SIZE
+   MADD  t4, a4, x1, y4
+   LD a4,  AO1,   7 * SIZE
+   LD y3,  YY,  10 * SIZE
+   LD y4,  YY,  11 * SIZE
+   ST t1,  YY,   0 * SIZE
+   ST t2,  YY,   1 * SIZE
+   ST t3,  YY,   2 * SIZE
+   ST t4,  YY,   3 * SIZE
+   MADD  t1, a1, x1, y5
+   LD a1,  AO1,   8 * SIZE
+   MADD  t2, a2, x1, y6
+   LD a2,  AO1,   9 * SIZE
+   LD y5,  YY,  12 * SIZE
+   LD y6,  YY,  13 * SIZE
+   MADD  t3, a3, x1, y7
+   LD a3,  AO1,  10 * SIZE
+   MADD  t4, a4, x1, y8
+   LD a4,  AO1,  11 * SIZE
+   LD y7,  YY,  14 * SIZE
+   LD y8,  YY,  15 * SIZE
+   ST t1,  YY,  4 * SIZE
+   ST t2,  YY,  5 * SIZE
+   ST t3,  YY,  6 * SIZE
+   ST t4,  YY,  7 * SIZE
+   addi.d  I, I, -1
+   addi.d  YY,  YY,   8 * SIZE
+   addi.d AO1, AO1,  8 * SIZE
+   blt $r0,    I, .L22
+   .align 3
+
+.L23:
+   MADD  t1, a1, x1, y1
+   LD a1,  AO1,   4 * SIZE
+   MADD  t2, a2, x1, y2
+   LD a2,  AO1,   5 * SIZE
+   MADD  t3, a3, x1, y3
+   LD a3,  AO1,   6 * SIZE
+   MADD  t4, a4, x1, y4
+   LD a4,  AO1,   7 * SIZE
+   ST t1,  YY,   0 * SIZE
+   MADD  t1, a1, x1, y5
+   ST t2,  YY,   1 * SIZE
+   MADD  t2, a2, x1, y6
+   ST t3,  YY,   2 * SIZE
+   MADD  t3, a3, x1, y7
+   ST t4,  YY,   3 * SIZE
+   MADD  t4, a4, x1, y8
+   ST t1,  YY,   4 * SIZE
+   ST t2,  YY,   5 * SIZE
+   ST t3,  YY,   6 * SIZE
+   ST t4,  YY,   7 * SIZE
+   addi.d  AO1, AO1,  8 * SIZE
+   addi.d  YY,  YY,   8 * SIZE
+   .align 3
+
+.L25:
+   andi    I,  M, 4
+   bge $r0,    I, .L26
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   LD y4,  YY,  3 * SIZE
+   MADD  y1, a1, x1, y1
+   MADD  y2, a2, x1, y2
+   MADD  y3, a3, x1, y3
+   addi.d  YY,  YY,   4 * SIZE
+   MADD  y4, a4, x1, y4
+   addi.d  AO1, AO1,  4 * SIZE
+   ST y1,  YY,  -4 * SIZE
+   ST y2,  YY,  -3 * SIZE
+   ST y3,  YY,  -2 * SIZE
+   ST y4,  YY,  -1 * SIZE
+   .align 3
+
+.L26:
+   andi    I,  M, 2
+   bge $r0,    I, .L27
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   MADD  y1, a1, x1, y1
+   addi.d  YY,  YY,   2 * SIZE
+   MADD  y2, a2, x1, y2
+   addi.d  AO1, AO1,  2 * SIZE
+   ST y1,  YY,  -2 * SIZE
+   ST y2,  YY,  -1 * SIZE
+   .align 3
+
+.L27:
+   andi    I,  M, 1
+   bge $r0,    I, .L900
+   LD y1,  YY,  0 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   MADD  y1, a1, x1, y1
+   ST y1,  YY,  0 * SIZE
+   .align 3
+
+.L900:
+   li  YORIG, SIZE
+   srai.d I,  M, 2
+   beq INCY, YORIG, .L999
+   move   XX, BUFFER
+   bge $r0,    I, .L905
+   .align 3
+
+.L902:
+   LD a1,  XX,  0 * SIZE
+   LD a2,  XX,  1 * SIZE
+   LD a3,  XX,  2 * SIZE
+   LD a4,  XX,  3 * SIZE
+   ST a1,  Y,  0 * SIZE
+   add.d   Y, Y, INCY
+   ST a2,  Y,  0 * SIZE
+   add.d   Y, Y, INCY
+   ST a3,  Y,  0 * SIZE
+   add.d   Y, Y, INCY
+   ST a4,  Y,  0 * SIZE
+   add.d   Y, Y, INCY
+   addi.d  I, I, -1
+   addi.d XX, XX, 4 * SIZE
+   blt $r0,    I, .L902
+   .align 3
+
+.L905:
+   andi    I,  M, 3
+   bge $r0,    I, .L999
+   .align 3
+
+.L906:
+   LD a1,  XX,  0 * SIZE
+   addi.d  XX, XX, 1 * SIZE
+   ST a1,  Y,  0 * SIZE
+   addi.d  I, I, -1
+   add.d  Y, Y, INCY
+   blt $r0,    I, .L906
+   .align 3
+
+.L999:
+   LDARG  $r23,  $sp,    0
+   LDARG  $r24,  $sp,    8
+#ifndef __64BIT__
+   fld.d  $f18,  $sp,  16
+   fld.d  $f19,  $sp,  24
+   fld.d  $f20,  $sp,  32
+#endif
+#ifdef __64BIT__
+   addi.d  $sp, $sp, 16
+#else
+   addi.d  $sp, $sp, 48
+#endif
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/gemv_t.S

@@ -0,0 +1,436 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+/* Unused param dummy1 */
+#define M      $r4
+#define N      $r5
+#define A      $r7
+#define LDA    $r8
+#define X      $r9
+#define INCX   $r10
+#define Y      $r11
+#define INCY   $r6
+#define BUFFER $r16
+#define XORIG  $r18
+#define XX     $r12
+#define YY     $r13
+#define I      $r14
+#define J      $r15
+#define AO1    $r23
+#define AO2    $r24
+#define ALPHA  $f0
+#define a1     $f22
+#define a2     $f8
+#define a3     $f23
+#define a4     $f9
+#define a5     $f10
+#define a6     $f11
+#define a7     $f12
+#define a8     $f13
+#define y1     $f14
+#define y2     $f15
+#define y3     $f16
+#define y4     $f17
+#define x1     $f3
+#define x2     $f1
+#define x3     $f2
+#define x4     $f4
+#define x5     $f5
+#define x6     $f6
+#define x7     $f7
+#define x8     $f18
+
+   PROLOGUE
+
+   LDARG  INCY,    $sp,  0
+   LDARG  BUFFER,  $sp,  8
+#ifdef __64BIT__
+   addi.d  $sp, $sp, -16
+#else
+   addi.d  $sp, $sp, -32
+#endif
+   MTC  y1, $r0
+   SDARG  $r23,  $sp,    0
+   SDARG  $r24,  $sp,    8
+   slli.d     LDA,     LDA,  BASE_SHIFT
+#ifndef __64BIT__
+   fst.d  $f18,  $sp,  16
+#endif
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    M, .L999
+   slli.d INCY, INCY, BASE_SHIFT
+   bge $r0,    N, .L999
+   li  I, SIZE
+   move   XORIG, X
+   beq INCX, I, .L10
+   srai.d  I,  M, 2
+   move    XORIG, BUFFER
+   move   YY, BUFFER
+   bge $r0,    I, .L05
+   .align 3
+
+.L02:
+   LD a1,  X,  0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,  0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,  0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,  0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  YY,  0 * SIZE
+   ST a2,  YY,  1 * SIZE
+   ST a3,  YY,  2 * SIZE
+   ST a4,  YY,  3 * SIZE
+   addi.d  I, I, -1
+   addi.d YY, YY, 4 * SIZE
+   blt $r0,    I, .L02
+   .align 3
+
+.L05:
+   andi    I,  M, 3
+   bge $r0,    I, .L10
+   .align 3
+
+.L06:
+   LD a1,  X,  0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  YY,  0 * SIZE
+   addi.d  I, I, -1
+   addi.d YY, YY, 1 * SIZE
+   blt $r0,    I, .L06
+   .align 3
+
+.L10:
+   srai.d  J,  N, 1
+   move   YY, Y
+   bge $r0,    J, .L20
+   .align 3
+
+.L11:
+   move    AO1, A
+   MOV y2, y1
+   add.d   AO2, A,      LDA
+   MOV y3, y1
+   add.d   A,   AO2,    LDA
+   MOV y4, y1
+   srai.d  I,  M, 3
+   move   XX, XORIG
+   bge $r0,    I, .L15
+   LD a1,  AO1,  0 * SIZE
+   LD x1,  XX,  0 * SIZE
+   LD a2,  AO2,  0 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD a3,  AO1,  1 * SIZE
+   LD x3,  XX,  2 * SIZE
+   LD a4,  AO2,  1 * SIZE
+   LD x4,  XX,  3 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   LD x5,  XX,  4 * SIZE
+   LD a6,  AO2,  2 * SIZE
+   LD x6,  XX,  5 * SIZE
+   LD a7,  AO1,  3 * SIZE
+   LD x7,  XX,  6 * SIZE
+   LD a8,  AO2,  3 * SIZE
+   addi.d  I, I, -1
+   LD x8,  XX,  7 * SIZE
+   bge $r0,    I, .L13
+   .align  3
+.L12:
+   MADD  y1, a1, x1, y1
+   LD a1,  AO1,  4 * SIZE
+   MADD  y2, a2, x1, y2
+   LD a2,  AO2,  4 * SIZE
+   MADD  y3, a3, x2, y3
+   LD a3,  AO1,  5 * SIZE
+   MADD  y4, a4, x2, y4
+   LD a4,  AO2,  5 * SIZE
+   LD x1,  XX,   8 * SIZE
+   LD x2,  XX,   9 * SIZE
+   MADD  y1, a5, x3, y1
+   LD a5,  AO1,  6 * SIZE
+   MADD  y2, a6, x3, y2
+   LD a6,  AO2,  6 * SIZE
+   MADD  y3, a7, x4, y3
+   LD a7,  AO1,  7 * SIZE
+   MADD  y4, a8, x4, y4
+   LD a8,  AO2,  7 * SIZE
+   LD x3,  XX,  10 * SIZE
+   LD x4,  XX,  11 * SIZE
+   MADD  y1, a1, x5, y1
+   LD a1,  AO1,  8 * SIZE
+   MADD  y2, a2, x5, y2
+   LD a2,  AO2,  8 * SIZE
+   MADD  y3, a3, x6, y3
+   LD a3,  AO1,  9 * SIZE
+   MADD  y4, a4, x6, y4
+   LD a4,  AO2,  9 * SIZE
+   LD x5,  XX,  12 * SIZE
+   LD x6,  XX,  13 * SIZE
+   MADD  y1, a5, x7, y1
+   LD a5,  AO1, 10 * SIZE
+   MADD  y2, a6, x7, y2
+   LD a6,  AO2, 10 * SIZE
+   MADD  y3, a7, x8, y3
+   LD a7,  AO1, 11 * SIZE
+   MADD  y4, a8, x8, y4
+   LD a8,  AO2, 11 * SIZE
+   LD x7,  XX,  14 * SIZE
+   LD x8,  XX,  15 * SIZE
+   addi.d  I, I, -1
+   addi.d  XX,  XX,   8 * SIZE
+   addi.d  AO1, AO1,  8 * SIZE
+   addi.d AO2, AO2,  8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   MADD  y1, a1, x1, y1
+   LD a1,  AO1,  4 * SIZE
+   MADD  y2, a2, x1, y2
+   LD a2,  AO2,  4 * SIZE
+   MADD  y3, a3, x2, y3
+   LD a3,  AO1,  5 * SIZE
+   MADD  y4, a4, x2, y4
+   LD a4,  AO2,  5 * SIZE
+   MADD  y1, a5, x3, y1
+   LD a5,  AO1,  6 * SIZE
+   MADD  y2, a6, x3, y2
+   LD a6,  AO2,  6 * SIZE
+   MADD  y3, a7, x4, y3
+   LD a7,  AO1,  7 * SIZE
+   MADD  y4, a8, x4, y4
+   LD a8,  AO2,  7 * SIZE
+   MADD  y1, a1, x5, y1
+   MADD  y2, a2, x5, y2
+   MADD  y3, a3, x6, y3
+   MADD  y4, a4, x6, y4
+   MADD  y1, a5, x7, y1
+   addi.d  XX,  XX,   8 * SIZE
+   MADD  y2, a6, x7, y2
+   addi.d  AO1, AO1,  8 * SIZE
+   MADD  y3, a7, x8, y3
+   addi.d  AO2, AO2,  8 * SIZE
+   MADD  y4, a8, x8, y4
+   .align 3
+
+.L15:
+   andi    I,  M, 4
+   bge $r0,    I, .L17
+   LD a1,  AO1,  0 * SIZE
+   LD x1,  XX,  0 * SIZE
+   LD a2,  AO2,  0 * SIZE
+   LD a3,  AO1,  1 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD a4,  AO2,  1 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   LD x3,  XX,  2 * SIZE
+   MADD  y1, a1, x1, y1
+   LD a6,  AO2,  2 * SIZE
+   MADD  y2, a2, x1, y2
+   LD a7,  AO1,  3 * SIZE
+   MADD  y3, a3, x2, y3
+   LD x4,  XX,  3 * SIZE
+   MADD  y4, a4, x2, y4
+   LD a8,  AO2,  3 * SIZE
+   MADD  y1, a5, x3, y1
+   MADD  y2, a6, x3, y2
+   addi.d  XX,  XX,   4 * SIZE
+   MADD  y3, a7, x4, y3
+   addi.d  AO1, AO1,  4 * SIZE
+   MADD  y4, a8, x4, y4
+   addi.d  AO2, AO2,  4 * SIZE
+   .align 3
+
+.L17:
+   andi    I,  M, 3
+   ADD y1, y1, y3
+   ADD    y2, y2, y4
+   bge $r0,    I, .L19
+   .align  3
+.L18:
+   LD x1,  XX,  0 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD a2,  AO2,  0 * SIZE
+   addi.d  I, I, -1
+   addi.d  XX, XX, 1 * SIZE
+   addi.d  AO1, AO1,  1 * SIZE
+   addi.d  AO2, AO2,  1 * SIZE
+   MADD  y1, a1, x1, y1
+   MADD  y2, a2, x1, y2
+   blt $r0,    I, .L18
+   .align 3
+
+.L19:
+   LD a1,  Y,  0 * SIZE
+   add.d   Y, Y, INCY
+   LD a2,  Y,  0 * SIZE
+   add.d   Y, Y, INCY
+   MADD  a1, y1, ALPHA, a1
+   addi.d  J, J, -1
+   MADD  a2, y2, ALPHA, a2
+   MTC  y1, $r0
+   ST a1,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST a2,  YY,   0 * SIZE
+   add.d  YY, YY, INCY
+   blt $r0,    J, .L11
+   .align 3
+
+.L20:
+   andi    J,  N, 1
+   MOV y3, y1
+   move   AO1, A
+   bge $r0,    J, .L999
+   srai.d  I,  M, 3
+   move   XX, XORIG
+   bge $r0,    I, .L25
+   LD a1,  AO1,  0 * SIZE
+   LD x1,  XX,  0 * SIZE
+   LD a3,  AO1,  1 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   LD x3,  XX,  2 * SIZE
+   LD a7,  AO1,  3 * SIZE
+   LD x4,  XX,  3 * SIZE
+   LD x5,  XX,  4 * SIZE
+   LD x6,  XX,  5 * SIZE
+   LD x7,  XX,  6 * SIZE
+   addi.d  I, I, -1
+   LD x8,  XX,  7 * SIZE
+   bge $r0,    I, .L23
+   .align  3
+.L22:
+   MADD  y1, a1, x1, y1
+   LD a1,  AO1,  4 * SIZE
+   MADD  y3, a3, x2, y3
+   LD a3,  AO1,  5 * SIZE
+   LD x1,  XX,   8 * SIZE
+   LD x2,  XX,   9 * SIZE
+   MADD  y1, a5, x3, y1
+   LD a5,  AO1,  6 * SIZE
+   MADD  y3, a7, x4, y3
+   LD a7,  AO1,  7 * SIZE
+   LD x3,  XX,  10 * SIZE
+   LD x4,  XX,  11 * SIZE
+   MADD  y1, a1, x5, y1
+   LD a1,  AO1,   8 * SIZE
+   MADD  y3, a3, x6, y3
+   LD a3,  AO1,   9 * SIZE
+   LD x5,  XX,  12 * SIZE
+   LD x6,  XX,  13 * SIZE
+   MADD  y1, a5, x7, y1
+   LD a5,  AO1,  10 * SIZE
+   MADD  y3, a7, x8, y3
+   LD a7,  AO1,  11 * SIZE
+   LD x7,  XX,  14 * SIZE
+   LD x8,  XX,  15 * SIZE
+   addi.d  I, I, -1
+   addi.d  XX,  XX,   8 * SIZE
+   addi.d AO1, AO1,  8 * SIZE
+   blt $r0,    I, .L22
+   .align 3
+
+.L23:
+   MADD  y1, a1, x1, y1
+   LD a1,  AO1,  4 * SIZE
+   MADD  y3, a3, x2, y3
+   LD a3,  AO1,  5 * SIZE
+   MADD  y1, a5, x3, y1
+   LD a5,  AO1,  6 * SIZE
+   MADD  y3, a7, x4, y3
+   LD a7,  AO1,  7 * SIZE
+   MADD  y1, a1, x5, y1
+   MADD  y3, a3, x6, y3
+   MADD  y1, a5, x7, y1
+   MADD  y3, a7, x8, y3
+   addi.d  XX,  XX,   8 * SIZE
+   addi.d  AO1, AO1,  8 * SIZE
+   .align 3
+
+.L25:
+   andi    I,  M, 4
+   bge $r0,    I, .L27
+   LD a1,  AO1,  0 * SIZE
+   LD x1,  XX,  0 * SIZE
+   LD a3,  AO1,  1 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   LD x3,  XX,  2 * SIZE
+   MADD  y1, a1, x1, y1
+   LD a7,  AO1,  3 * SIZE
+   MADD  y3, a3, x2, y3
+   LD x4,  XX,  3 * SIZE
+   MADD  y1, a5, x3, y1
+   addi.d  XX,  XX,   4 * SIZE
+   MADD  y3, a7, x4, y3
+   addi.d  AO1, AO1,  4 * SIZE
+   .align 3
+
+.L27:
+   andi    I,  M, 3
+   ADD y1, y1, y3
+   bge $r0,    I, .L29
+   .align  3
+.L28:
+   LD x1,  XX,  0 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   addi.d  I, I, -1
+   addi.d  XX, XX, 1 * SIZE
+   addi.d  AO1, AO1,  1 * SIZE
+   MADD  y1, a1, x1, y1
+   blt $r0,    I, .L28
+   .align 3
+
+.L29:
+   LD a1,  Y,  0 * SIZE
+   add.d   Y, Y, INCY
+   MADD  a1, y1, ALPHA, a1
+   ST a1,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   .align 3
+
+.L999:
+   LDARG  $r23,  $sp,    0
+   LDARG  $r24,  $sp,    8
+#ifndef __64BIT__
+   fld.d  $f18,  $sp,  16
+#endif
+#ifdef __64BIT__
+   addi.d  $sp, $sp, 16
+#else
+   addi.d  $sp, $sp, 32
+#endif
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/iamax.S

@@ -0,0 +1,233 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r18
+#define TEMP   $r7
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+#define x1     $r17
+#define x2     $r8
+#define x3     $r9
+#define x4     $r10
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   li x1, 0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD a1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   li x1, 1
+   bge $r0,    N, .L999
+   FABS    s1, a1
+   add.d   X, X, INCX
+   FABS    s2, a1
+   li  x2, 1
+   FABS    s3, a1
+   srai.d  I, N, 3
+   FABS    s4, a1
+   li  x3, 1
+   li  TEMP, 2
+   li x4, 1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+   FABS    t3, a3
+   LD a2,  X,   0 * SIZE
+   FABS    t4, a4
+   add.d   X, X, INCX
+   CMPLT   $fcc0, s1, t1
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, s3, t3
+   LD a4,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, t4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d  TEMP, TEMP, 4
+   addi.d  I, I, -1
+   FABS    t1, a5
+   LD a5,  X,   0 * SIZE
+   FABS    t2, a6
+   add.d   X, X, INCX
+   FABS    t3, a7
+   LD a6,  X,   0 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+   CMPLT   $fcc0, s1, t1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, s3, t3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, t4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d TEMP, TEMP, 4
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMPLT   $fcc2, s3, t3
+   CMPLT   $fcc3, s4, t4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   FABS    t1, a5
+   addi.d  TEMP, TEMP, 4
+   FABS    t2, a6
+   FABS    t3, a7
+   FABS    t4, a8
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMPLT   $fcc2, s3, t3
+   CMPLT   $fcc3, s4, t4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d  TEMP, TEMP, 4
+   addi.d  x2, x2, 1
+   addi.d  x3, x3, 2
+   addi.d  x4, x4, 3
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   addi.d  I, I, -1
+   CMPLT   $fcc0, s1, t1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   addi.d TEMP, TEMP, 1
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s1, s2
+   CMPLT   $fcc1, s3, s4
+   CMOVT  s1,  s1,  s2,  $fcc0
+   MOVT(x1, x2, $fcc0)
+   CMOVT  s3,  s3,  s4,  $fcc1
+   MOVT(x3, x4, $fcc1)
+   CMPLT   $fcc0, s1, s3
+   CMOVT  s1,  s1,  s3,  $fcc0
+   MOVT(x1, x3, $fcc0)
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/iamin.S

@@ -0,0 +1,233 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r18
+#define TEMP   $r7
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+#define x1     $r17
+#define x2     $r8
+#define x3     $r9
+#define x4     $r10
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   li x1, 0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD a1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   li x1, 1
+   bge $r0,    N, .L999
+   FABS    s1, a1
+   add.d   X, X, INCX
+   FABS    s2, a1
+   li  x2, 1
+   FABS    s3, a1
+   srai.d  I, N, 3
+   FABS    s4, a1
+   li  x3, 1
+   li  TEMP, 2
+   li x4, 1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+   FABS    t3, a3
+   LD a2,  X,   0 * SIZE
+   FABS    t4, a4
+   add.d   X, X, INCX
+   CMPLT   $fcc0, t1, s1
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, t2, s2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, t3, s3
+   LD a4,  X,   0 * SIZE
+   CMPLT   $fcc3, t4, s4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d  TEMP, TEMP, 4
+   addi.d  I, I, -1
+   FABS    t1, a5
+   LD a5,  X,   0 * SIZE
+   FABS    t2, a6
+   add.d   X, X, INCX
+   FABS    t3, a7
+   LD a6,  X,   0 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+   CMPLT   $fcc0, t1, s1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, t2, s2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, t3, s3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, t4, s4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d TEMP, TEMP, 4
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   CMPLT   $fcc0, t1, s1
+   CMPLT   $fcc1, t2, s2
+   CMPLT   $fcc2, t3, s3
+   CMPLT   $fcc3, t4, s4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   FABS    t1, a5
+   addi.d  TEMP, TEMP, 4
+   FABS    t2, a6
+   FABS    t3, a7
+   FABS    t4, a8
+   CMPLT   $fcc0, t1, s1
+   CMPLT   $fcc1, t2, s2
+   CMPLT   $fcc2, t3, s3
+   CMPLT   $fcc3, t4, s4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d  TEMP, TEMP, 4
+   addi.d  x2, x2, 1
+   addi.d  x3, x3, 2
+   addi.d  x4, x4, 3
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   addi.d  I, I, -1
+   CMPLT   $fcc0, t1, s1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   addi.d TEMP, TEMP, 1
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s2, s1
+   CMPLT   $fcc1, s4, s3
+   CMOVT  s1,  s1,  s2,  $fcc0
+   MOVT(x1, x2, $fcc0)
+   CMOVT  s3,  s3,  s4,  $fcc1
+   MOVT(x3, x4, $fcc1)
+   CMPLT   $fcc0, s3, s1
+   CMOVT  s1,  s1,  s3,  $fcc0
+   MOVT(x1, x3, $fcc0)
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/izamax.S

@@ -0,0 +1,217 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r18
+#define TEMP   $r7
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define t5     $f4
+#define t6     $f5
+#define t7     $f6
+#define t8     $f7
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+#define x1     $r17
+#define x2     $r8
+#define x3     $r9
+#define x4     $r10
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+      LDINT   N,     0(N)
+      LDINT   INCX,  0(INCX)
+#endif
+
+   li x1, 0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD s1, t1, t2
+   ADD s2, t1, t2
+   ADD s3, t1, t2
+   ADD s4, t1, t2
+   addi.d  N, N, -1
+   li x1, 1
+   bge $r0,    N, .L999
+   add.d   X, X, INCX
+   li  x2, 1
+   srai.d  I, N, 2
+   li  x3, 1
+   li  TEMP, 2
+   li x4, 1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   LD a8,  X,   1 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   LD a2,  X,   1 * SIZE
+   FABS    t3, a3
+   add.d   X, X, INCX
+   FABS    t4, a4
+   FABS    t5, a5
+   LD a3,  X,   0 * SIZE
+   FABS    t6, a6
+   LD a4,  X,   1 * SIZE
+   FABS    t7, a7
+   add.d   X, X, INCX
+   FABS    t8, a8
+   ADD t1, t1, t2
+   LD a5,  X,   0 * SIZE
+   ADD t3, t3, t4
+   LD a6,  X,   1 * SIZE
+   ADD t5, t5, t6
+   add.d   X, X, INCX
+   ADD t7, t7, t8
+   CMPLT   $fcc0, s1, t1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t3
+   LD a8,  X,   1 * SIZE
+   CMPLT   $fcc2, s3, t5
+   add.d   X, X, INCX
+   CMPLT   $fcc3, s4, t7
+   addi.d  I, I, -1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t3,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t5,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t7,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d TEMP, TEMP, 4
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   FABS    t5, a5
+   FABS    t6, a6
+   FABS    t7, a7
+   FABS    t8, a8
+   ADD t1, t1, t2
+   ADD t3, t3, t4
+   ADD t5, t5, t6
+   ADD t7, t7, t8
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t3
+   CMPLT   $fcc2, s3, t5
+   CMPLT   $fcc3, s4, t7
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t3,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t5,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t7,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d  TEMP, TEMP, 4
+   addi.d  x2, x2, 1
+   addi.d  x3, x3, 2
+   addi.d  x4, x4, 3
+   .align 3
+
+.L15:
+   andi    I,  N, 3
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD t1, t1, t2
+   addi.d  I, I, -1
+   CMPLT   $fcc0, s1, t1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   addi.d TEMP, TEMP, 1
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s1, s2
+   CMPLT   $fcc1, s3, s4
+   CMOVT  s1,  s1,  s2,  $fcc0
+   MOVT(x1, x2, $fcc0)
+   CMOVT  s3,  s3,  s4,  $fcc1
+   MOVT(x3, x4, $fcc1)
+   CMPLT   $fcc0, s1, s3
+   CMOVT  s1,  s1,  s3,  $fcc0
+   MOVT(x1, x3, $fcc0)
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/izamin.S

@@ -0,0 +1,217 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r18
+#define TEMP   $r7
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define t5     $f4
+#define t6     $f5
+#define t7     $f6
+#define t8     $f7
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+#define x1     $r17
+#define x2     $r8
+#define x3     $r9
+#define x4     $r10
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   li x1, 0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD s1, t1, t2
+   ADD s2, t1, t2
+   ADD s3, t1, t2
+   ADD s4, t1, t2
+   addi.d  N, N, -1
+   li x1, 1
+   bge $r0,    N, .L999
+   add.d   X, X, INCX
+   li  x2, 1
+   srai.d  I, N, 2
+   li  x3, 1
+   li  TEMP, 2
+   li x4, 1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   LD a8,  X,   1 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   LD a2,  X,   1 * SIZE
+   FABS    t3, a3
+   add.d   X, X, INCX
+   FABS    t4, a4
+   FABS    t5, a5
+   LD a3,  X,   0 * SIZE
+   FABS    t6, a6
+   LD a4,  X,   1 * SIZE
+   FABS    t7, a7
+   add.d   X, X, INCX
+   FABS    t8, a8
+   ADD t1, t1, t2
+   LD a5,  X,   0 * SIZE
+   ADD t3, t3, t4
+   LD a6,  X,   1 * SIZE
+   ADD t5, t5, t6
+   add.d   X, X, INCX
+   ADD t7, t7, t8
+   CMPLT   $fcc0, t1, s1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, t3, s2
+   LD a8,  X,   1 * SIZE
+   CMPLT   $fcc2, t5, s3
+   add.d   X, X, INCX
+   CMPLT   $fcc3, t7, s4
+   addi.d  I, I, -1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t3,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t5,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t7,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d TEMP, TEMP, 4
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   FABS    t5, a5
+   FABS    t6, a6
+   FABS    t7, a7
+   FABS    t8, a8
+   ADD t1, t1, t2
+   ADD t3, t3, t4
+   ADD t5, t5, t6
+   ADD t7, t7, t8
+   CMPLT   $fcc0, t1, s1
+   CMPLT   $fcc1, t3, s2
+   CMPLT   $fcc2, t5, s3
+   CMPLT   $fcc3, t7, s4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t3,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t5,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t7,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d  TEMP, TEMP, 4
+   addi.d  x2, x2, 1
+   addi.d  x3, x3, 2
+   addi.d  x4, x4, 3
+   .align 3
+
+.L15:
+   andi    I,  N, 3
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD t1, t1, t2
+   addi.d  I, I, -1
+   CMPLT   $fcc0, t1, s1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   addi.d TEMP, TEMP, 1
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s2, s1
+   CMPLT   $fcc1, s4, s3
+   CMOVT  s1,  s1,  s2,  $fcc0
+   MOVT(x1, x2, $fcc0)
+   CMOVT  s3,  s3,  s4,  $fcc1
+   MOVT(x3, x4, $fcc1)
+   CMPLT   $fcc0, s3, s1
+   CMOVT  s1,  s1,  s3,  $fcc0
+   MOVT(x1, x3, $fcc0)
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/max.S

@@ -0,0 +1,174 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+      LDINT   N,     0(N)
+      LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD s1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   add.d   X, X, INCX
+   MOV    s2, s1
+   bge $r0,    N, .L999
+   MOV s3, s1
+   srai.d  I, N, 3
+   MOV    s4, s1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   CMPLT   $fcc0, s1, a1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, a2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, s3, a3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, a4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  a1,  $fcc0
+   LD a1,  X,   0 * SIZE
+   CMOVT  s2,  s2,  a2,  $fcc1
+   add.d   X, X, INCX
+   CMOVT  s3,  s3,  a3,  $fcc2
+   LD a2,  X,   0 * SIZE
+   CMOVT  s4,  s4,  a4,  $fcc3
+   add.d   X, X, INCX
+   CMPLT   $fcc0, s1, a5
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, a6
+   add.d   X, X, INCX
+   CMPLT   $fcc2, s3, a7
+   LD a4,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, a8
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  a5,  $fcc0
+   LD a5,  X,   0 * SIZE
+   CMOVT  s2,  s2,  a6,  $fcc1
+   add.d   X, X, INCX
+   CMOVT  s3,  s3,  a7,  $fcc2
+   LD a6,  X,   0 * SIZE
+   CMOVT  s4,  s4,  a8,  $fcc3
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   CMPLT   $fcc0, s1, a1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, a2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, s3, a3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, a4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  a1,  $fcc0
+   CMOVT  s2,  s2,  a2,  $fcc1
+   CMOVT  s3,  s3,  a3,  $fcc2
+   CMOVT  s4,  s4,  a4,  $fcc3
+   CMPLT   $fcc0, s1, a5
+   CMPLT   $fcc1, s2, a6
+   CMPLT   $fcc2, s3, a7
+   CMPLT   $fcc3, s4, a8
+   CMOVT  s1,  s1,  a5,  $fcc0
+   CMOVT  s2,  s2,  a6,  $fcc1
+   CMOVT  s3,  s3,  a7,  $fcc2
+   CMOVT  s4,  s4,  a8,  $fcc3
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   CMPLT   $fcc0, s1, a1
+   CMOVT  s1,  s1,  a1,  $fcc0
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s1, s2
+   CMPLT   $fcc1, s3, s4
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+   CMPLT   $fcc0, s1, s3
+   CMOVT  s1,  s1,  s3,  $fcc0
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/min.S

@@ -0,0 +1,174 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD s1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   add.d   X, X, INCX
+   MOV    s2, s1
+   bge $r0,    N, .L999
+   MOV s3, s1
+   srai.d  I, N, 3
+   MOV    s4, s1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   CMPLT   $fcc0, a1, s1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, a2, s2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, a3, s3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, a4, s4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  a1,  $fcc0
+   LD a1,  X,   0 * SIZE
+   CMOVT  s2,  s2,  a2,  $fcc1
+   add.d   X, X, INCX
+   CMOVT  s3,  s3,  a3,  $fcc2
+   LD a2,  X,   0 * SIZE
+   CMOVT  s4,  s4,  a4,  $fcc3
+   add.d   X, X, INCX
+   CMPLT   $fcc0, a5, s1
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, a6, s2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, a7, s3
+   LD a4,  X,   0 * SIZE
+   CMPLT   $fcc3, a8, s4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  a5,  $fcc0
+   LD a5,  X,   0 * SIZE
+   CMOVT  s2,  s2,  a6,  $fcc1
+   add.d   X, X, INCX
+   CMOVT  s3,  s3,  a7,  $fcc2
+   LD a6,  X,   0 * SIZE
+   CMOVT  s4,  s4,  a8,  $fcc3
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   CMPLT   $fcc0, a1, s1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, a2, s2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, a3, s3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, a4, s4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  a1,  $fcc0
+   CMOVT  s2,  s2,  a2,  $fcc1
+   CMOVT  s3,  s3,  a3,  $fcc2
+   CMOVT  s4,  s4,  a4,  $fcc3
+   CMPLT   $fcc0, a5, s1
+   CMPLT   $fcc1, a6, s2
+   CMPLT   $fcc2, a7, s3
+   CMPLT   $fcc3, a8, s4
+   CMOVT  s1,  s1,  a5,  $fcc0
+   CMOVT  s2,  s2,  a6,  $fcc1
+   CMOVT  s3,  s3,  a7,  $fcc2
+   CMOVT  s4,  s4,  a8,  $fcc3
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   CMPLT   $fcc0, a1, s1
+   CMOVT  s1,  s1,  a1,  $fcc0
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s2, s1
+   CMPLT   $fcc1, s4, s3
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+   CMPLT   $fcc0, s3, s1
+   CMOVT  s1,  s1,  s3,  $fcc0
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/scal.S

@@ -0,0 +1,330 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r7
+#define INCX   $r8
+
+#define I      $r17
+#define TEMP   $r18
+#define XX     $r5
+#define ALPHA  $f0
+#define a1     $f22
+#define a2     $f8
+#define a3     $f23
+#define a4     $f9
+#define a5     $f10
+#define a6     $f11
+#define a7     $f12
+#define a8     $f13
+#define t1     $f14
+#define t2     $f15
+#define t3     $f16
+#define t4     $f17
+
+   PROLOGUE
+
+   li  TEMP, SIZE
+   MTC  a1, $r0
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    N, .L999
+   CMPEQ   $fcc0, ALPHA, a1
+   bceqz   $fcc0, .L50
+   srai.d I, N, 3
+   bne INCX, TEMP, .L20
+   bge $r0,    I, .L15
+   .align 3
+
+.L12:
+   ST a1,  X,   0 * SIZE
+   ST a1,  X,   1 * SIZE
+   ST a1,  X,   2 * SIZE
+   ST a1,  X,   3 * SIZE
+   ST a1,  X,   4 * SIZE
+   ST a1,  X,   5 * SIZE
+   ST a1,  X,   6 * SIZE
+   ST a1,  X,   7 * SIZE
+   addi.w  I, I, -1
+   addi.d X, X, 8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L16:
+   ST a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   addi.d X, X, SIZE
+   blt $r0,    I, .L16
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   .align 3
+
+.L20:
+   srai.d  I, N, 3
+   bge $r0,    I, .L25
+   .align 3
+
+.L22:
+   ST a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L22
+   .align 3
+
+.L25:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L26:
+   addi.d  I, I, -1
+   ST a1,  X,   0 * SIZE
+   add.d  X, X, INCX
+   blt $r0,    I, .L26
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   .align 3
+
+.L50:
+   srai.d I, N, 3
+   bne INCX, TEMP, .L60
+   addi.d I, I, -1
+   blt I,  $r0, .L55
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   LD a5,  X,   4 * SIZE
+   LD a6,  X,   5 * SIZE
+   LD a7,  X,   6 * SIZE
+   LD a8,  X,   7 * SIZE
+   bge $r0,    I, .L53
+   .align 3
+
+.L52:
+   MUL t1, ALPHA, a1
+   LD a1,  X,   8 * SIZE
+   MUL t2, ALPHA, a2
+   LD a2,  X,   9 * SIZE
+   MUL t3, ALPHA, a3
+   LD a3,  X,  10 * SIZE
+   MUL t4, ALPHA, a4
+   LD a4,  X,  11 * SIZE
+   ST t1,  X,   0 * SIZE
+   MUL t1, ALPHA, a5
+   LD a5,  X,  12 * SIZE
+   ST t2,  X,   1 * SIZE
+   MUL t2, ALPHA, a6
+   LD a6,  X,  13 * SIZE
+   ST t3,  X,   2 * SIZE
+   MUL t3, ALPHA, a7
+   LD a7,  X,  14 * SIZE
+   ST t4,  X,   3 * SIZE
+   MUL t4, ALPHA, a8
+   LD a8,  X,  15 * SIZE
+   addi.d  I, I, -1
+   ST t1,  X,   4 * SIZE
+   ST t2,  X,   5 * SIZE
+   ST t3,  X,   6 * SIZE
+   ST t4,  X,   7 * SIZE
+   addi.d X, X, 8 * SIZE
+   blt $r0,    I, .L52
+   .align 3
+
+.L53:
+   MUL t1, ALPHA, a1
+   MUL t2, ALPHA, a2
+   MUL t3, ALPHA, a3
+   MUL t4, ALPHA, a4
+   ST t1,  X,   0 * SIZE
+   MUL t1, ALPHA, a5
+   ST t2,  X,   1 * SIZE
+   MUL t2, ALPHA, a6
+   ST t3,  X,   2 * SIZE
+   MUL t3, ALPHA, a7
+   ST t4,  X,   3 * SIZE
+   MUL t4, ALPHA, a8
+   ST t1,  X,   4 * SIZE
+   ST t2,  X,   5 * SIZE
+   ST t3,  X,   6 * SIZE
+   ST t4,  X,   7 * SIZE
+   addi.d  X, X, 8 * SIZE
+   .align 3
+
+.L55:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L56:
+   LD a1,  X,   0 * SIZE
+   MUL t1, ALPHA, a1
+   addi.d  X, X, SIZE
+   addi.d  I, I, -1
+   ST t1,  X,  -1 * SIZE
+   blt $r0,    I, .L56
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   .align 3
+
+.L60:
+   srai.d  I, N, 3
+   move    XX, X
+   addi.d I, I, -1
+   blt I,  $r0, .L65
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   add.d   X, X, INCX
+   bge $r0,    I, .L63
+   .align 3
+
+.L62:
+   MUL t1, ALPHA, a1
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   MUL t2, ALPHA, a2
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   MUL t3, ALPHA, a3
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   MUL t4, ALPHA, a4
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST t1,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t2,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t3,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t4,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   MUL t1, ALPHA, a5
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   MUL t2, ALPHA, a6
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   MUL t3, ALPHA, a7
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   MUL t4, ALPHA, a8
+   LD a8,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST t1,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t2,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t3,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t4,  XX,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  XX, XX, INCX
+   blt $r0,    I, .L62
+   .align 3
+
+.L63:
+   MUL t1, ALPHA, a1
+   MUL t2, ALPHA, a2
+   MUL t3, ALPHA, a3
+   MUL t4, ALPHA, a4
+   ST t1,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t2,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t3,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t4,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   MUL t1, ALPHA, a5
+   MUL t2, ALPHA, a6
+   MUL t3, ALPHA, a7
+   MUL t4, ALPHA, a8
+   ST t1,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t2,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t3,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t4,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   .align 3
+
+.L65:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L66:
+   LD a1,  X,   0 * SIZE
+   MUL t1, ALPHA, a1
+   addi.d  I, I, -1
+   ST t1,  X,   0 * SIZE
+   add.d  X, X, INCX
+   blt $r0,    I, .L66
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/snrm2.S

@@ -0,0 +1,249 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f12
+#define a2     $f13
+#define a3     $f14
+#define a4     $f15
+#define a5     $f16
+#define a6     $f17
+#define a7     $f0
+#define a8     $f1
+#define s1     $f22
+#define s2     $f8
+#define t1     $f23
+#define t2     $f9
+#define t3     $f10
+#define t4     $f11
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   movgr2fr.d  s1, $r0
+   li  TEMP, SIZE
+   fmov.d s2, s1
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+   srai.d I, N, 3
+   bne INCX, TEMP, .L20
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   LD a5,  X,   4 * SIZE
+   addi.d  I, I, -1
+   fcvt.d.s    t1, a1
+   LD a6,  X,   5 * SIZE
+   fcvt.d.s    t2, a2
+   LD a7,  X,   6 * SIZE
+   fcvt.d.s    t3, a3
+   LD a8,  X,   7 * SIZE
+   fcvt.d.s   t4, a4
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   fmadd.d  s1, t1, t1, s1
+   LD a1,  X,   8 * SIZE
+   fcvt.d.s    t1, a5
+   NOP
+   fmadd.d  s2, t2, t2, s2
+   LD a2,  X,   9 * SIZE
+   fcvt.d.s    t2, a6
+   NOP
+   fmadd.d  s1, t3, t3, s1
+   LD a3,  X,  10 * SIZE
+   fcvt.d.s    t3, a7
+   NOP
+   fmadd.d  s2, t4, t4, s2
+   LD a4,  X,  11 * SIZE
+   fcvt.d.s    t4, a8
+   NOP
+   fmadd.d  s1, t1, t1, s1
+   LD a5,  X,  12 * SIZE
+   fcvt.d.s    t1, a1
+   NOP
+   fmadd.d  s2, t2, t2, s2
+   LD a6,  X,  13 * SIZE
+   fcvt.d.s    t2, a2
+   addi.d  I, I, -1
+   fmadd.d  s1, t3, t3, s1
+   LD a7,  X,  14 * SIZE
+   fcvt.d.s    t3, a3
+   addi.d  X, X, 8 * SIZE
+   fmadd.d  s2, t4, t4, s2
+   LD a8,  X,   7 * SIZE
+   fcvt.d.s   t4, a4
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   fmadd.d  s1, t1, t1, s1
+   fcvt.d.s    t1, a5
+   fmadd.d  s2, t2, t2, s2
+   fcvt.d.s    t2, a6
+   fmadd.d  s1, t3, t3, s1
+   fcvt.d.s    t3, a7
+   fmadd.d  s2, t4, t4, s2
+   fcvt.d.s    t4, a8
+   fmadd.d  s1, t1, t1, s1
+   fmadd.d  s2, t2, t2, s2
+   fmadd.d  s1, t3, t3, s1
+   fmadd.d  s2, t4, t4, s2
+   addi.d  X, X, 8 * SIZE
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   fcvt.d.s    t1, a1
+   fmadd.d  s1, t1, t1, s1
+   addi.d X, X, SIZE
+   blt $r0,    I, .L16
+   b   .L999
+   .align 3
+
+.L20:
+   bge $r0,    I, .L25
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   addi.d  I, I, -1
+   fcvt.d.s    t1, a1
+   fcvt.d.s    t2, a2
+   fcvt.d.s    t3, a3
+   fcvt.d.s    t4, a4
+   add.d  X, X, INCX
+   bge $r0,    I, .L24
+   .align 3
+
+.L23:
+   fmadd.d  s1, t1, t1, s1
+   LD a1,  X,   0 * SIZE
+   fcvt.d.s    t1, a5
+   add.d   X, X, INCX
+   fmadd.d  s2, t2, t2, s2
+   LD a2,  X,   0 * SIZE
+   fcvt.d.s    t2, a6
+   add.d   X, X, INCX
+   fmadd.d  s1, t3, t3, s1
+   LD a3,  X,   0 * SIZE
+   fcvt.d.s    t3, a7
+   add.d   X, X, INCX
+   fmadd.d  s2, t4, t4, s2
+   LD a4,  X,   0 * SIZE
+   fcvt.d.s    t4, a8
+   add.d   X, X, INCX
+   fmadd.d  s1, t1, t1, s1
+   LD a5,  X,   0 * SIZE
+   fcvt.d.s    t1, a1
+   add.d   X, X, INCX
+   fmadd.d  s2, t2, t2, s2
+   LD a6,  X,   0 * SIZE
+   fcvt.d.s    t2, a2
+   add.d   X, X, INCX
+   fmadd.d  s1, t3, t3, s1
+   LD a7,  X,   0 * SIZE
+   fcvt.d.s    t3, a3
+   add.d   X, X, INCX
+   fmadd.d  s2, t4, t4, s2
+   LD a8,  X,   0 * SIZE
+   fcvt.d.s    t4, a4
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L23
+   .align 3
+
+.L24:
+   fmadd.d  s1, t1, t1, s1
+   fcvt.d.s    t1, a5
+   fmadd.d  s2, t2, t2, s2
+   fcvt.d.s    t2, a6
+   fmadd.d  s1, t3, t3, s1
+   fcvt.d.s    t3, a7
+   fmadd.d  s2, t4, t4, s2
+   fcvt.d.s    t4, a8
+   fmadd.d  s1, t1, t1, s1
+   fmadd.d  s2, t2, t2, s2
+   fmadd.d  s1, t3, t3, s1
+   fmadd.d  s2, t4, t4, s2
+   .align 3
+
+.L25:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+
+.L26:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   fcvt.d.s    t1, a1
+   add.d   X, X, INCX
+   fmadd.d  s1, t1, t1, s1
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+   fadd.d  s1, s1, s2
+   fsqrt.d s1, s1
+   move $r4, $r17
+   fcvt.s.d    $f0, s1
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/swap.S

@@ -0,0 +1,330 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r7
+#define INCX   $r8
+#define Y      $r9
+#define INCY   $r10
+
+#define I      $r17
+#define TEMP   $r18
+#define XX     $r5
+#define YY     $r6
+#define a1     $f22
+#define a2     $f8
+#define a3     $f23
+#define a4     $f9
+#define a5     $f10
+#define a6     $f11
+#define a7     $f12
+#define a8     $f13
+#define b1     $f14
+#define b2     $f15
+#define b3     $f16
+#define b4     $f17
+#define b5     $f0
+#define b6     $f1
+#define b7     $f2
+#define b8     $f3
+
+   PROLOGUE
+
+   li  TEMP, SIZE
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    N, .L999
+   slli.d INCY, INCY, BASE_SHIFT
+   bne INCX, TEMP, .L20
+   srai.d I, N, 3
+   bne INCY, TEMP, .L20
+   addi.d I, I, -1
+   blt I,  $r0, .L15
+   LD a1,  X,   0 * SIZE
+   LD b1,  Y,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD b2,  Y,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD b3,  Y,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   LD b4,  Y,   3 * SIZE
+   LD a5,  X,   4 * SIZE
+   LD b5,  Y,   4 * SIZE
+   LD a6,  X,   5 * SIZE
+   LD b6,  Y,   5 * SIZE
+   LD a7,  X,   6 * SIZE
+   LD b7,  Y,   6 * SIZE
+   LD a8,  X,   7 * SIZE
+   LD b8,  Y,   7 * SIZE
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   ST a1,  Y,   0 * SIZE
+   LD a1,  X,   8 * SIZE
+   ST b1,  X,   0 * SIZE
+   LD b1,  Y,   8 * SIZE
+   ST a2,  Y,   1 * SIZE
+   LD a2,  X,   9 * SIZE
+   ST b2,  X,   1 * SIZE
+   LD b2,  Y,   9 * SIZE
+   ST a3,  Y,   2 * SIZE
+   LD a3,  X,  10 * SIZE
+   ST b3,  X,   2 * SIZE
+   LD b3,  Y,  10 * SIZE
+   ST a4,  Y,   3 * SIZE
+   LD a4,  X,  11 * SIZE
+   ST b4,  X,   3 * SIZE
+   LD b4,  Y,  11 * SIZE
+   ST a5,  Y,   4 * SIZE
+   LD a5,  X,  12 * SIZE
+   ST b5,  X,   4 * SIZE
+   LD b5,  Y,  12 * SIZE
+   ST a6,  Y,   5 * SIZE
+   LD a6,  X,  13 * SIZE
+   ST b6,  X,   5 * SIZE
+   LD b6,  Y,  13 * SIZE
+   ST a7,  Y,   6 * SIZE
+   LD a7,  X,  14 * SIZE
+   ST b7,  X,   6 * SIZE
+   LD b7,  Y,  14 * SIZE
+   ST a8,  Y,   7 * SIZE
+   LD a8,  X,  15 * SIZE
+   ST b8,  X,   7 * SIZE
+   LD b8,  Y,  15 * SIZE
+   addi.d  I, I, -1
+   addi.d  X, X, 8 * SIZE
+   addi.d Y, Y, 8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   ST a1,  Y,   0 * SIZE
+   ST b1,  X,   0 * SIZE
+   ST a2,  Y,   1 * SIZE
+   ST b2,  X,   1 * SIZE
+   ST a3,  Y,   2 * SIZE
+   ST b3,  X,   2 * SIZE
+   ST a4,  Y,   3 * SIZE
+   ST b4,  X,   3 * SIZE
+   ST a5,  Y,   4 * SIZE
+   ST b5,  X,   4 * SIZE
+   ST a6,  Y,   5 * SIZE
+   ST b6,  X,   5 * SIZE
+   ST a7,  Y,   6 * SIZE
+   ST b7,  X,   6 * SIZE
+   ST a8,  Y,   7 * SIZE
+   ST b8,  X,   7 * SIZE
+   addi.d  X, X, 8 * SIZE
+   addi.d  Y, Y, 8 * SIZE
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD b1,  Y,   0 * SIZE
+   addi.d  X, X, SIZE
+   addi.d  I, I, -1
+   addi.d  Y, Y, SIZE
+   ST b1,  X,  -1 * SIZE
+   ST a1,  Y,  -1 * SIZE
+   blt $r0,    I, .L16
+   b   .L999
+   .align 3
+
+.L20:
+   srai.d  I, N, 3
+   move    XX, X
+   move    YY, Y
+   addi.d I, I, -1
+   blt I,  $r0, .L25
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b2,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b3,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b4,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b5,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b6,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b7,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a8,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b8,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   bge $r0,    I, .L23
+   .align 3
+
+.L22:
+   ST a1,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b1,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a2,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b2,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b2,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a3,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b3,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b3,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a4,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b4,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b4,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a5,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b5,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b5,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a6,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b6,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b6,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a7,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b7,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b7,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a8,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a8,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b8,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b8,  Y,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  Y, Y, INCY
+   blt $r0,    I, .L22
+   .align 3
+
+.L23:
+   ST a1,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b1,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST a2,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b2,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST a3,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b3,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST a4,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b4,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST a5,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b5,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST a6,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b6,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST a7,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b7,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST a8,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b8,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   .align 3
+
+.L25:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L26:
+   LD a1,  X,   0 * SIZE
+   LD b1,  Y,   0 * SIZE
+   addi.d  I, I, -1
+   ST a1,  Y,   0 * SIZE
+   ST b1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   add.d  Y, Y, INCY
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/zamax.S

@@ -0,0 +1,190 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define t5     $f4
+#define t6     $f5
+#define t7     $f6
+#define t8     $f7
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD a1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD    s1, t1, t2
+   bge $r0,    N, .L999
+   ADD s2, t1, t2
+   srai.d  I, N, 2
+   ADD s3, t1, t2
+   ADD    s4, t1, t2
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   LD a8,  X,   1 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   LD a2,  X,   1 * SIZE
+   FABS    t3, a3
+   add.d   X, X, INCX
+   FABS    t4, a4
+   FABS    t5, a5
+   LD a3,  X,   0 * SIZE
+   FABS    t6, a6
+   LD a4,  X,   1 * SIZE
+   FABS    t7, a7
+   add.d   X, X, INCX
+   FABS    t8, a8
+   ADD t1, t1, t2
+   LD a5,  X,   0 * SIZE
+   ADD t3, t3, t4
+   LD a6,  X,   1 * SIZE
+   ADD t5, t5, t6
+   add.d   X, X, INCX
+   ADD t7, t7, t8
+   CMPLT   $fcc0, s1, t1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t3
+   LD a8,  X,   1 * SIZE
+   CMPLT   $fcc2, s3, t5
+   add.d   X, X, INCX
+   CMPLT   $fcc3, s4, t7
+   CMOVT  s1,  s1,  t1,  $fcc0
+   addi.d  I, I, -1
+   CMOVT  s2,  s2,  t3,  $fcc1
+   CMOVT  s3,  s3,  t5,  $fcc2
+   CMOVT  s4,  s4,  t7,  $fcc3
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   FABS    t5, a5
+   FABS    t6, a6
+   FABS    t7, a7
+   FABS    t8, a8
+   ADD t1, t1, t2
+   ADD t3, t3, t4
+   ADD t5, t5, t6
+   ADD t7, t7, t8
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t3
+   CMPLT   $fcc2, s3, t5
+   CMPLT   $fcc3, s4, t7
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t3,  $fcc1
+   CMOVT  s3,  s3,  t5,  $fcc2
+   CMOVT  s4,  s4,  t7,  $fcc3
+   .align 3
+
+.L15:
+   andi    I,  N, 3
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   addi.d  I, I, -1
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD t1, t1, t2
+   CMPLT   $fcc0, s1, t1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s1, s2
+   CMPLT   $fcc1, s3, s4
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+   CMPLT   $fcc0, s1, s3
+   CMOVT  s1,  s1,  s3,  $fcc0
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/zamin.S

@@ -0,0 +1,198 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N     $r4
+#define X     $r5
+#define INCX  $r6
+#define I     $r17
+#define TEMP  $r18
+#define a1    $f10
+#define a2    $f11
+#define a3    $f12
+#define a4    $f13
+#define a5    $f14
+#define a6    $f15
+#define a7    $f16
+#define a8    $f17
+#define t1    $f0
+#define t2    $f1
+#define t3    $f2
+#define t4    $f3
+#define t5    $f4
+#define t6    $f5
+#define t7    $f6
+#define t8    $f7
+#define s1    $f22
+#define s2    $f8
+#define s3    $f23
+#define s4    $f9
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD a1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD    s1, t1, t2
+   bge $r0,    N, .L999
+   NOP
+   ADD s2, t1, t2
+   srai.d  I, N, 2
+   ADD s3, t1, t2
+   ADD    s4, t1, t2
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   LD a8,  X,   1 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   LD a2,  X,   1 * SIZE
+   FABS    t3, a3
+   add.d   X, X, INCX
+   FABS    t4, a4
+   NOP
+   FABS    t5, a5
+   LD a3,  X,   0 * SIZE
+   FABS    t6, a6
+   LD a4,  X,   1 * SIZE
+   FABS    t7, a7
+   add.d   X, X, INCX
+   FABS    t8, a8
+   NOP
+   ADD t1, t1, t2
+   LD a5,  X,   0 * SIZE
+   ADD t3, t3, t4
+   LD a6,  X,   1 * SIZE
+   ADD t5, t5, t6
+   add.d   X, X, INCX
+   ADD t7, t7, t8
+   NOP
+   CMPLT   $fcc0, t1, s1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, t3, s2
+   LD a8,  X,   1 * SIZE
+   CMPLT   $fcc2, t5, s3
+   add.d   X, X, INCX
+   CMPLT   $fcc3, t7, s4
+   NOP
+   CMOVT  s1,  s1,  t1,  $fcc0
+   addi.d  I, I, -1
+   CMOVT  s2,  s2,  t3,  $fcc1
+   NOP
+   CMOVT  s3,  s3,  t5,  $fcc2
+   CMOVT  s4,  s4,  t7,  $fcc3
+   blt $r0,    I, .L12
+   NOP
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   FABS    t5, a5
+   FABS    t6, a6
+   FABS    t7, a7
+   FABS    t8, a8
+   ADD t1, t1, t2
+   ADD t3, t3, t4
+   ADD t5, t5, t6
+   ADD t7, t7, t8
+   CMPLT   $fcc0, t1, s1
+   CMPLT   $fcc1, t3, s2
+   CMPLT   $fcc2, t5, s3
+   CMPLT   $fcc3, t7, s4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t3,  $fcc1
+   CMOVT  s3,  s3,  t5,  $fcc2
+   CMOVT  s4,  s4,  t7,  $fcc3
+   .align 3
+
+.L15:
+   andi    I,  N, 3
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   addi.d  I, I, -1
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD t1, t1, t2
+   CMPLT   $fcc0, t1, s1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s2, s1
+   CMPLT   $fcc1, s4, s3
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+   CMPLT   $fcc0, s3, s1
+   CMOVT  s1,  s1,  s3,  $fcc0
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   NOP
+
+   EPILOGUE

kernel/loongarch64/zasum.S

@@ -0,0 +1,158 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f23
+#define a2     $f9
+#define a3     $f10
+#define a4     $f11
+#define a5     $f12
+#define a6     $f13
+#define a7     $f14
+#define a8     $f15
+#define t1     $f16
+#define t2     $f17
+#define t3     $f0
+#define t4     $f1
+#define s1     $f22
+#define s2     $f8
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   MTC  s2, $r0
+   slli.d  INCX, INCX, ZBASE_SHIFT
+   srai.d I, N, 2
+   bge $r0,    N, .L999
+   bge $r0,    I, .L25
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   FABS    t2, a2
+   LD a7,  X,   0 * SIZE
+   LD a8,  X,   1 * SIZE
+   FABS    t3, a3
+   FABS    t4, a4
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L24
+   .align 3
+
+.L23:
+   ADD s1, s1, t1
+   LD a1,  X,   0 * SIZE
+   FABS    t1, a5
+   addi.d  I, I, -1
+   ADD s2, s2, t2
+   LD a2,  X,   1 * SIZE
+   FABS    t2, a6
+   add.d   X, X, INCX
+   ADD s1, s1, t3
+   LD a3,  X,   0 * SIZE
+   FABS    t3, a7
+   NOP
+   ADD s2, s2, t4
+   LD a4,  X,   1 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+   ADD s1, s1, t1
+   LD a5,  X,   0 * SIZE
+   FABS    t1, a1
+   NOP
+   ADD s2, s2, t2
+   LD a6,  X,   1 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+   ADD s1, s1, t3
+   LD a7,  X,   0 * SIZE
+   FABS    t3, a3
+   LD a8,  X,   1 * SIZE
+   ADD s2, s2, t4
+   add.d   X, X, INCX
+   FABS   t4, a4
+   blt $r0,    I, .L23
+   .align 3
+
+.L24:
+   ADD s1, s1, t1
+   FABS    t1, a5
+   ADD s2, s2, t2
+   FABS    t2, a6
+   ADD s1, s1, t3
+   FABS    t3, a7
+   ADD s2, s2, t4
+   FABS    t4, a8
+   ADD s1, s1, t1
+   ADD s2, s2, t2
+   ADD s1, s1, t3
+   ADD s2, s2, t4
+   .align 3
+
+.L25:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   .align  3
+
+.L26:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   FABS    t1, a1
+   addi.d  I, I, -1
+   FABS    t2, a2
+   add.d   X, X, INCX
+   ADD s1, s1, t1
+   ADD    s2, s2, t2
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+   ADD s1, s1, s2
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/zcopy.S

@@ -0,0 +1,217 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define Y      $r7
+#define INCY   $r8
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f22
+#define a2     $f8
+#define a3     $f23
+#define a4     $f9
+#define a5     $f10
+#define a6     $f11
+#define a7     $f12
+#define a8     $f13
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+   LDINT   INCY,  0(INCY)
+#endif
+
+   li  TEMP, 2 * SIZE
+   NOP
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    N, .L999
+   slli.d INCY, INCY, ZBASE_SHIFT
+   bne INCX, TEMP, .L20
+   srai.d I, N, 2
+   bne INCY, TEMP, .L20
+   addi.d I, I, -1
+   blt     I, $r0,     .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   LD a5,  X,   4 * SIZE
+   LD a6,  X,   5 * SIZE
+   LD a7,  X,   6 * SIZE
+   LD a8,  X,   7 * SIZE
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   ST a1,  Y,   0 * SIZE
+   LD a1,  X,   8 * SIZE
+   ST a2,  Y,   1 * SIZE
+   LD a2,  X,   9 * SIZE
+   ST a3,  Y,   2 * SIZE
+   LD a3,  X,  10 * SIZE
+   ST a4,  Y,   3 * SIZE
+   LD a4,  X,  11 * SIZE
+   ST a5,  Y,   4 * SIZE
+   LD a5,  X,  12 * SIZE
+   ST a6,  Y,   5 * SIZE
+   LD a6,  X,  13 * SIZE
+   ST a7,  Y,   6 * SIZE
+   LD a7,  X,  14 * SIZE
+   ST a8,  Y,   7 * SIZE
+   LD a8,  X,  15 * SIZE
+   addi.d  I, I, -1
+   addi.d  X, X, 8 * SIZE
+   addi.d Y, Y, 8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   ST a1,  Y,   0 * SIZE
+   ST a2,  Y,   1 * SIZE
+   ST a3,  Y,   2 * SIZE
+   ST a4,  Y,   3 * SIZE
+   ST a5,  Y,   4 * SIZE
+   ST a6,  Y,   5 * SIZE
+   ST a7,  Y,   6 * SIZE
+   ST a8,  Y,   7 * SIZE
+   addi.d  X, X, 8 * SIZE
+   addi.d  Y, Y, 8 * SIZE
+   .align 3
+
+.L15:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   addi.d  X, X, 2 * SIZE
+   addi.d  Y, Y, 2 * SIZE
+   ST a1,  Y,  -2 * SIZE
+   addi.d  I, I, -1
+   ST a2,  Y,  -1 * SIZE
+   blt $r0,    I, .L16
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   NOP
+   .align 3
+
+.L20:
+   srai.d  I, N, 2
+   addi.d I, I, -1
+   blt I,  $r0, .L25
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   LD a8,  X,   1 * SIZE
+   add.d  X, X, INCX
+   bge $r0,    I, .L23
+   .align 3
+
+.L22:
+   ST a1,  Y,   0 * SIZE
+   LD a1,  X,   0 * SIZE
+   ST a2,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   ST a3,  Y,   0 * SIZE
+   LD a3,  X,   0 * SIZE
+   ST a4,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   ST a5,  Y,   0 * SIZE
+   LD a5,  X,   0 * SIZE
+   ST a6,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   ST a7,  Y,   0 * SIZE
+   LD a7,  X,   0 * SIZE
+   ST a8,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   LD a8,  X,   1 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L22
+   .align 3
+
+.L23:
+   ST a1,  Y,   0 * SIZE
+   ST a2,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   ST a3,  Y,   0 * SIZE
+   ST a4,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   ST a5,  Y,   0 * SIZE
+   ST a6,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   ST a7,  Y,   0 * SIZE
+   ST a8,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   .align 3
+
+.L25:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   .align  3
+
+.L26:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   addi.d  I, I, -1
+   ST a1,  Y,   0 * SIZE
+   ST a2,  Y,   1 * SIZE
+   add.d  Y, Y, INCY
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/zdot.S

@@ -0,0 +1,330 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define Y      $r7
+#define INCY   $r8
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define b1     $f14
+#define b2     $f15
+#define b3     $f16
+#define b4     $f17
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+   LDINT   INCY,  0(INCY)
+#endif
+
+   MTC  s1, $r0
+   MOV s2, s1
+   MOV s3, s2
+   MOV s4, s3
+   slli.d  INCX, INCX, ZBASE_SHIFT
+   li  TEMP, 2 * SIZE
+   slli.d INCY, INCY, ZBASE_SHIFT
+   bge $r0,    N, .L999
+   srai.d I, N, 2
+   bne INCX, TEMP, .L20
+   bne INCY, TEMP, .L20
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD b1,  Y,   0 * SIZE
+   addi.d  I, I, -1
+   LD b2,  Y,   1 * SIZE
+   bge $r0,    I, .L14
+   .align 3
+
+.L13:
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   2 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   3 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   2 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   3 * SIZE
+   MADD  s1, b3, a3, s1
+   LD a1,  X,   4 * SIZE
+   MADD  s2, b3, a4, s2
+   LD a2,  X,   5 * SIZE
+   MADD  s3, b4, a3, s3
+   LD b1,  Y,   4 * SIZE
+   MADD  s4, b4, a4, s4
+   LD b2,  Y,   5 * SIZE
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   6 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   7 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   6 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   7 * SIZE
+   MADD  s1, b3, a3, s1
+   LD a1,  X,   8 * SIZE
+   MADD  s2, b3, a4, s2
+   LD a2,  X,   9 * SIZE
+   MADD  s3, b4, a3, s3
+   LD b1,  Y,   8 * SIZE
+   MADD  s4, b4, a4, s4
+   LD b2,  Y,   9 * SIZE
+   addi.d  I, I, -1
+   addi.d  X, X, 8 * SIZE
+   addi.d Y, Y, 8 * SIZE
+   blt $r0,    I, .L13
+   .align 3
+
+.L14:
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   2 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   3 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   2 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   3 * SIZE
+   MADD  s1, b3, a3, s1
+   LD a1,  X,   4 * SIZE
+   MADD  s2, b3, a4, s2
+   LD a2,  X,   5 * SIZE
+   MADD  s3, b4, a3, s3
+   LD b1,  Y,   4 * SIZE
+   MADD  s4, b4, a4, s4
+   LD b2,  Y,   5 * SIZE
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   6 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   7 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   6 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   7 * SIZE
+   MADD  s1, b3, a3, s1
+   addi.d  X, X, 8 * SIZE
+   MADD  s2, b3, a4, s2
+   addi.d  Y, Y, 8 * SIZE
+   MADD  s3, b4, a3, s3
+   MADD  s4, b4, a4, s4
+   .align 3
+
+.L15:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD b1,  Y,   0 * SIZE
+   addi.d  I, I, -1
+   LD b2,  Y,   1 * SIZE
+   bge $r0,    I, .L17
+   .align 3
+
+.L16:
+   MADD  s1, b1, a1, s1
+   addi.d  I, I, -1
+   MADD  s2, b1, a2, s2
+   LD b1,  Y,   2 * SIZE
+   MADD  s3, b2, a1, s3
+   LD a1,  X,   2 * SIZE
+   MADD  s4, b2, a2, s4
+   LD a2,  X,   3 * SIZE
+   LD b2,  Y,   3 * SIZE
+   addi.d  X, X, 2 * SIZE
+   addi.d Y, Y, 2 * SIZE
+   blt $r0,    I, .L16
+   .align 3
+
+.L17:
+   MADD  s1, b1, a1, s1
+   MADD  s2, b1, a2, s2
+   MADD  s3, b2, a1, s3
+   MADD  s4, b2, a2, s4
+   b   .L999
+   .align 3
+
+.L20:
+#ifdef F_INTERFACE
+   bgez    INCX, .L21
+   addi.d  TEMP, N, -1
+   mult    TEMP, INCX
+   mflo    TEMP
+   dsub    X, X, TEMP
+   .align 3
+
+.L21:
+   bgez    INCY, .L22
+   addi.d  TEMP, N, -1
+   mult    TEMP, INCY
+   mflo    TEMP
+   dsub    Y, Y, TEMP
+   .align 3
+
+.L22:
+#endif
+   bge $r0,    I, .L25
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD b1,  Y,   0 * SIZE
+   LD b2,  Y,   1 * SIZE
+   add.d   X, X, INCX
+   addi.d  I, I, -1
+   add.d  Y, Y, INCY
+   bge $r0,    I, .L24
+   .align 3
+
+.L23:
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   0 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   1 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   0 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   1 * SIZE
+   add.d   X, X, INCX
+   add.d   Y, Y, INCY
+   MADD  s1, b3, a3, s1
+   LD a1,  X,   0 * SIZE
+   MADD  s2, b3, a4, s2
+   LD a2,  X,   1 * SIZE
+   MADD  s3, b4, a3, s3
+   LD b1,  Y,   0 * SIZE
+   MADD  s4, b4, a4, s4
+   LD b2,  Y,   1 * SIZE
+   add.d   X, X, INCX
+   add.d   Y, Y, INCY
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   0 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   1 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   0 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   1 * SIZE
+   add.d   X, X, INCX
+   add.d   Y, Y, INCY
+   MADD  s1, b3, a3, s1
+   LD a1,  X,   0 * SIZE
+   MADD  s2, b3, a4, s2
+   LD a2,  X,   1 * SIZE
+   MADD  s3, b4, a3, s3
+   LD b1,  Y,   0 * SIZE
+   MADD  s4, b4, a4, s4
+   LD b2,  Y,   1 * SIZE
+   add.d   X, X, INCX
+   addi.d  I, I, -1
+   add.d  Y, Y, INCY
+   blt $r0,    I, .L23
+   .align 3
+
+.L24:
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   0 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   1 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   0 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   1 * SIZE
+   add.d   X, X, INCX
+   add.d   Y, Y, INCY
+   MADD  s1, b3, a3, s1
+   LD a1,  X,   0 * SIZE
+   MADD  s2, b3, a4, s2
+   LD a2,  X,   1 * SIZE
+   MADD  s3, b4, a3, s3
+   LD b1,  Y,   0 * SIZE
+   MADD  s4, b4, a4, s4
+   LD b2,  Y,   1 * SIZE
+   add.d   X, X, INCX
+   add.d   Y, Y, INCY
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   0 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   1 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   0 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   1 * SIZE
+   MADD  s1, b3, a3, s1
+   add.d   X, X, INCX
+   MADD  s2, b3, a4, s2
+   add.d   Y, Y, INCY
+   MADD  s3, b4, a3, s3
+   MADD  s4, b4, a4, s4
+   .align 3
+
+.L25:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   .align  3
+.L26:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD b1,  Y,   0 * SIZE
+   LD b2,  Y,   1 * SIZE
+   MADD  s1, b1, a1, s1
+   MADD  s2, b1, a2, s2
+   MADD  s3, b2, a1, s3
+   MADD  s4, b2, a2, s4
+   add.d   X, X, INCX
+   add.d   Y, Y, INCY
+   addi.d  I, I, -1
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+#ifndef CONJ
+   SUB $f0, s1, s4
+#else
+   ADD $f0, s1, s4
+#endif
+#ifndef CONJ
+   ADD $f1, s3, s2
+#else
+   SUB $f1, s3, s2
+#endif
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/zgemv_n.S

@@ -0,0 +1,648 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define M          $r4
+#define N          $r5
+#define A          $r7
+#define LDA        $r8
+#define X          $r9
+#define INCX       $r10
+#define Y          $r11
+#define INCY       $r6
+#define BUFFER     $r17
+
+#define YORIG      $r18
+#define XX         $r12
+#define YY         $r13
+#define I          $r14
+#define J          $r15
+#define AO1        $r23
+#define AO2        $r24
+
+#define ALPHA_R    $f0
+#define ALPHA_I    $f1
+#define a1         $f22
+#define a2         $f8
+#define a3         $f23
+#define a4         $f9
+#define a5         $f10
+#define a6         $f11
+#define a7         $f12
+#define a8         $f13
+#define x1         $f14
+#define x2         $f15
+#define x3         $f16
+#define x4         $f17
+#define y1         $f3
+#define y2         $f4
+#define y3         $f2
+#define y4         $f5
+#define t1         $f6
+#define t2         $f7
+#define t3         $f18
+#define t4         $f19
+#define t5         $f20
+#define t6         $f21
+#define t7         $f24
+#define t8         $f25
+
+#if !defined(CONJ) && !defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        MADD
+#define    MADD3        NMSUB
+#define    MADD4        MADD
+#endif
+#if  defined(CONJ) && !defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        MADD
+#define    MADD3        MADD
+#define    MADD4        NMSUB
+#endif
+#if  !defined(CONJ) && defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        NMSUB
+#define    MADD3        MADD
+#define    MADD4        MADD
+#endif
+#if  defined(CONJ) && defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        NMSUB
+#define    MADD3        NMSUB
+#define    MADD4        NMSUB
+#endif
+
+   PROLOGUE
+
+   LDARG  INCY,    $sp,  0
+   LDARG  BUFFER,  $sp,  8
+#ifndef __64BIT__
+   addi.d  $sp, $sp, -64
+#else
+   addi.d  $sp, $sp, -32
+#endif
+   SDARG  $r23,  $sp,    0
+   SDARG  $r24,  $sp,    8
+   fst.d  $f24,  $sp,  16
+   fst.d  $f25,  $sp,  24
+#ifndef __64BIT__
+   fst.d  $f18,  $sp,  32
+   fst.d  $f19,  $sp,  40
+   fst.d  $f20,  $sp,  48
+   fst.d  $f21,  $sp,  56
+#endif
+   slli.d     LDA,     LDA,  ZBASE_SHIFT
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    M, .L999
+   slli.d INCY, INCY, ZBASE_SHIFT
+   bge $r0,    N, .L999
+   li  I, 2 * SIZE
+   move   YORIG, Y
+   beq INCY, I, .L10
+   srai.d  I,  M, 2
+   move    YORIG, BUFFER
+   move    XX, Y
+   move   YY, BUFFER
+   bge $r0,    I, .L05
+   .align 3
+
+.L02:
+   LD a1,  XX,  0 * SIZE
+   LD a2,  XX,  1 * SIZE
+   add.d   XX, XX, INCY
+   LD a3,  XX,  0 * SIZE
+   LD a4,  XX,  1 * SIZE
+   add.d   XX, XX, INCY
+   LD a5,  XX,  0 * SIZE
+   LD a6,  XX,  1 * SIZE
+   add.d   XX, XX, INCY
+   LD a7,  XX,  0 * SIZE
+   LD a8,  XX,  1 * SIZE
+   add.d   XX, XX, INCY
+   addi.d  I, I, -1
+   addi.d  YY, YY, 8 * SIZE
+   ST a1,  YY,  -8 * SIZE
+   ST a2,  YY,  -7 * SIZE
+   ST a3,  YY,  -6 * SIZE
+   ST a4,  YY,  -5 * SIZE
+   ST a5,  YY,  -4 * SIZE
+   ST a6,  YY,  -3 * SIZE
+   ST a7,  YY,  -2 * SIZE
+   ST a8,  YY,  -1 * SIZE
+   blt $r0,    I, .L02
+   .align 3
+
+.L05:
+   andi    I,  M, 3
+   bge $r0,    I, .L10
+   .align 3
+
+.L06:
+   LD a1,  XX,  0 * SIZE
+   LD a2,  XX,  1 * SIZE
+   add.d   XX, XX, INCY
+   addi.d  I, I, -1
+   ST a1,  YY,  0 * SIZE
+   ST a2,  YY,  1 * SIZE
+   addi.d YY, YY, 2 * SIZE
+   blt $r0,    I, .L06
+   .align 3
+
+.L10:
+   srai.d  J,  N, 1
+   bge $r0,    J, .L20
+   .align 3
+
+.L11:
+   LD x1,  X,  0 * SIZE
+   LD x2,  X,  1 * SIZE
+   add.d   X, X, INCX
+   LD x3,  X,  0 * SIZE
+   LD x4,  X,  1 * SIZE
+   add.d   X, X, INCX
+   MUL a1, ALPHA_R, x1
+   move    AO1, A
+   MUL a2, ALPHA_I, x1
+   add.d   AO2, A,      LDA
+   MUL a3, ALPHA_R, x3
+   add.d   A,   AO2,    LDA
+   MUL a4, ALPHA_I, x3
+#ifndef XCONJ
+   NMSUB  x1, x2, ALPHA_I, a1
+   MADD  x2, x2, ALPHA_R, a2
+   NMSUB  x3, x4, ALPHA_I, a3
+   MADD  x4, x4, ALPHA_R, a4
+#else
+   MADD  x1, x2, ALPHA_I, a1
+   MSUB  x2, x2, ALPHA_R, a2
+   MADD  x3, x4, ALPHA_I, a3
+   MSUB  x4, x4, ALPHA_R, a4
+#endif
+   srai.d  I,  M, 2
+   move   YY, YORIG
+   bge $r0,    I, .L15
+   LD y1,  YY,  0 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y4,  YY,  3 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   LD a5,  AO2,  0 * SIZE
+   LD a6,  AO2,  1 * SIZE
+   LD a7,  AO2,  2 * SIZE
+   LD a8,  AO2,  3 * SIZE
+   MADD1  t1, a1, x1, y1
+   LD y1,  YY,   4 * SIZE
+   MADD2  t2, a1, x2, y2
+   LD a1,  AO1,   4 * SIZE
+   MADD1  t3, a3, x1, y3
+   LD y2,  YY,   5 * SIZE
+   MADD2  t4, a3, x2, y4
+   LD a3,  AO1,   6 * SIZE
+   MADD3  t1, a2, x2, t1
+   LD y3,  YY,   6 * SIZE
+   MADD4  t2, a2, x1, t2
+   LD a2,  AO1,   5 * SIZE
+   MADD3  t3, a4, x2, t3
+   LD y4,  YY,   7 * SIZE
+   MADD4  t4, a4, x1, t4
+   LD a4,  AO1,   7 * SIZE
+   MADD1  t1, a5, x3, t1
+   MADD2  t2, a5, x4, t2
+   LD a5,  AO2,   4 * SIZE
+   MADD1  t3, a7, x3, t3
+   MADD2  t4, a7, x4, t4
+   LD a7,  AO2,   6 * SIZE
+   MADD3  t1, a6, x4, t1
+   MADD4  t2, a6, x3, t2
+   LD a6,  AO2,   5 * SIZE
+   MADD3  t3, a8, x4, t3
+   addi.d  I, I, -1
+   MADD4  t4, a8, x3, t4
+   LD a8,  AO2,   7 * SIZE
+   bge $r0,    I, .L13
+   .align  3
+.L12:
+   MADD1  t5, a1, x1, y1
+   LD y1,  YY,   8 * SIZE
+   MADD2  t6, a1, x2, y2
+   LD a1,  AO1,   8 * SIZE
+   MADD1  t7, a3, x1, y3
+   LD y2,  YY,   9 * SIZE
+   MADD2  t8, a3, x2, y4
+   LD a3,  AO1,  10 * SIZE
+   MADD3  t5, a2, x2, t5
+   LD y3,  YY,  10 * SIZE
+   MADD4  t6, a2, x1, t6
+   LD a2,  AO1,   9 * SIZE
+   MADD3  t7, a4, x2, t7
+   LD y4,  YY,  11 * SIZE
+   MADD4  t8, a4, x1, t8
+   LD a4,  AO1,  11 * SIZE
+   MADD1  t5, a5, x3, t5
+   ST t1,  YY,   0 * SIZE
+   MADD2  t6, a5, x4, t6
+   LD a5,  AO2,   8 * SIZE
+   MADD1  t7, a7, x3, t7
+   ST t2,  YY,   1 * SIZE
+   MADD2  t8, a7, x4, t8
+   LD a7,  AO2,  10 * SIZE
+   MADD3  t5, a6, x4, t5
+   ST t3,  YY,   2 * SIZE
+   MADD4  t6, a6, x3, t6
+   LD a6,  AO2,   9 * SIZE
+   MADD3  t7, a8, x4, t7
+   ST t4,  YY,   3 * SIZE
+   MADD4  t8, a8, x3, t8
+   LD a8,  AO2,  11 * SIZE
+   MADD1  t1, a1, x1, y1
+   LD y1,  YY,  12 * SIZE
+   MADD2  t2, a1, x2, y2
+   LD a1,  AO1,  12 * SIZE
+   MADD1  t3, a3, x1, y3
+   LD y2,  YY,  13 * SIZE
+   MADD2  t4, a3, x2, y4
+   LD a3,  AO1,  14 * SIZE
+   MADD3  t1, a2, x2, t1
+   LD y3,  YY,  14 * SIZE
+   MADD4  t2, a2, x1, t2
+   LD a2,  AO1,  13 * SIZE
+   MADD3  t3, a4, x2, t3
+   LD y4,  YY,  15 * SIZE
+   MADD4  t4, a4, x1, t4
+   LD a4,  AO1,  15 * SIZE
+   MADD1  t1, a5, x3, t1
+   ST t5,  YY,   4 * SIZE
+   MADD2  t2, a5, x4, t2
+   LD a5,  AO2,  12 * SIZE
+   MADD1  t3, a7, x3, t3
+   ST t6,  YY,   5 * SIZE
+   MADD2  t4, a7, x4, t4
+   LD a7,  AO2,  14 * SIZE
+   MADD3  t1, a6, x4, t1
+   ST t7,  YY,   6 * SIZE
+   MADD4  t2, a6, x3, t2
+   LD a6,  AO2,  13 * SIZE
+   MADD3  t3, a8, x4, t3
+   ST t8,  YY,   7 * SIZE
+   MADD4  t4, a8, x3, t4
+   LD a8,  AO2,  15 * SIZE
+   addi.d  I, I, -1
+   addi.d  YY,  YY,   8 * SIZE
+   addi.d  AO1, AO1,  8 * SIZE
+   addi.d AO2, AO2,  8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   ST t1,  YY,   0 * SIZE
+   MADD1  t1, a1, x1, y1
+   ST t2,  YY,   1 * SIZE
+   MADD2  t2, a1, x2, y2
+   ST t3,  YY,   2 * SIZE
+   MADD1  t3, a3, x1, y3
+   ST t4,  YY,   3 * SIZE
+   MADD2  t4, a3, x2, y4
+   MADD3  t1, a2, x2, t1
+   MADD4  t2, a2, x1, t2
+   MADD3  t3, a4, x2, t3
+   MADD4  t4, a4, x1, t4
+   MADD1  t1, a5, x3, t1
+   MADD2  t2, a5, x4, t2
+   MADD1  t3, a7, x3, t3
+   MADD2  t4, a7, x4, t4
+   MADD3  t1, a6, x4, t1
+   addi.d  AO1, AO1,  8 * SIZE
+   MADD4  t2, a6, x3, t2
+   addi.d  AO2, AO2,  8 * SIZE
+   MADD3  t3, a8, x4, t3
+   addi.d  YY,  YY,   8 * SIZE
+   MADD4  t4, a8, x3, t4
+   ST t1,  YY,  -4 * SIZE
+   ST t2,  YY,  -3 * SIZE
+   ST t3,  YY,  -2 * SIZE
+   ST t4,  YY,  -1 * SIZE
+   .align 3
+
+.L15:
+   andi    I,  M, 2
+   bge $r0,    I, .L16
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   LD y4,  YY,  3 * SIZE
+   MADD1  t1, a1, x1, y1
+   LD a5,  AO2,  0 * SIZE
+   MADD2  t2, a1, x2, y2
+   LD a6,  AO2,  1 * SIZE
+   MADD1  t3, a3, x1, y3
+   LD a7,  AO2,  2 * SIZE
+   MADD2  t4, a3, x2, y4
+   LD a8,  AO2,  3 * SIZE
+   MADD3  t1, a2, x2, t1
+   MADD4  t2, a2, x1, t2
+   MADD3  t3, a4, x2, t3
+   MADD4  t4, a4, x1, t4
+   MADD1  t1, a5, x3, t1
+   MADD2  t2, a5, x4, t2
+   MADD1  t3, a7, x3, t3
+   MADD2  t4, a7, x4, t4
+   MADD3  t1, a6, x4, t1
+   addi.d  YY,  YY,   4 * SIZE
+   MADD4  t2, a6, x3, t2
+   addi.d  AO1, AO1,  4 * SIZE
+   MADD3  t3, a8, x4, t3
+   addi.d  AO2, AO2,  4 * SIZE
+   MADD4  t4, a8, x3, t4
+   ST t1,  YY,  -4 * SIZE
+   ST t2,  YY,  -3 * SIZE
+   ST t3,  YY,  -2 * SIZE
+   ST t4,  YY,  -1 * SIZE
+   .align 3
+
+.L16:
+   andi    I,  M, 1
+   bge $r0,    I, .L19
+   LD y1,  YY,  0 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   MADD1  t1, a1, x1, y1
+   LD a5,  AO2,  0 * SIZE
+   MADD2  t2, a1, x2, y2
+   LD a6,  AO2,  1 * SIZE
+   MADD3  t1, a2, x2, t1
+   MADD4  t2, a2, x1, t2
+   MADD1  t1, a5, x3, t1
+   MADD2  t2, a5, x4, t2
+   MADD3  t1, a6, x4, t1
+   MADD4  t2, a6, x3, t2
+   ST t1,  YY,   0 * SIZE
+   ST t2,  YY,   1 * SIZE
+   .align 3
+
+.L19:
+   addi.d  J, J, -1
+   blt $r0,    J, .L11
+   .align 3
+
+.L20:
+   andi    J,  N, 1
+   bge $r0,    J, .L900
+   LD x1,  X,  0 * SIZE
+   LD x2,  X,  1 * SIZE
+   add.d   X, X, INCX
+   MUL a1, ALPHA_R, x1
+   move    AO1, A
+   MUL a2, ALPHA_I, x1
+#ifndef XCONJ
+   NMSUB  x1, x2, ALPHA_I, a1
+   MADD  x2, x2, ALPHA_R, a2
+#else
+   MADD  x1, x2, ALPHA_I, a1
+   MSUB  x2, x2, ALPHA_R, a2
+#endif
+   srai.d  I,  M, 2
+   move   YY, YORIG
+   bge $r0,    I, .L25
+   LD y1,  YY,  0 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y4,  YY,  3 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   MADD1  t1, a1, x1, y1
+   LD y1,  YY,   4 * SIZE
+   MADD2  t2, a1, x2, y2
+   LD a1,  AO1,   4 * SIZE
+   MADD1  t3, a3, x1, y3
+   LD y2,  YY,   5 * SIZE
+   MADD2  t4, a3, x2, y4
+   LD a3,  AO1,   6 * SIZE
+   MADD3  t1, a2, x2, t1
+   LD y3,  YY,   6 * SIZE
+   MADD4  t2, a2, x1, t2
+   LD a2,  AO1,   5 * SIZE
+   MADD3  t3, a4, x2, t3
+   LD y4,  YY,   7 * SIZE
+   MADD4  t4, a4, x1, t4
+   addi.d  I, I, -1
+   LD a4,  AO1,   7 * SIZE
+   bge $r0,    I, .L23
+   .align  3
+.L22:
+   MADD1  t5, a1, x1, y1
+   LD y1,  YY,   8 * SIZE
+   MADD2  t6, a1, x2, y2
+   LD a1,  AO1,   8 * SIZE
+   MADD1  t7, a3, x1, y3
+   LD y2,  YY,   9 * SIZE
+   MADD2  t8, a3, x2, y4
+   LD a3,  AO1,  10 * SIZE
+   MADD3  t5, a2, x2, t5
+   LD y3,  YY,  10 * SIZE
+   MADD4  t6, a2, x1, t6
+   LD a2,  AO1,   9 * SIZE
+   MADD3  t7, a4, x2, t7
+   LD y4,  YY,  11 * SIZE
+   MADD4  t8, a4, x1, t8
+   LD a4,  AO1,  11 * SIZE
+   ST t1,  YY,   0 * SIZE
+   ST t2,  YY,   1 * SIZE
+   ST t3,  YY,   2 * SIZE
+   ST t4,  YY,   3 * SIZE
+   MADD1  t1, a1, x1, y1
+   LD y1,  YY,  12 * SIZE
+   MADD2  t2, a1, x2, y2
+   LD a1,  AO1,  12 * SIZE
+   MADD1  t3, a3, x1, y3
+   LD y2,  YY,  13 * SIZE
+   MADD2  t4, a3, x2, y4
+   LD a3,  AO1,  14 * SIZE
+   MADD3  t1, a2, x2, t1
+   LD y3,  YY,  14 * SIZE
+   MADD4  t2, a2, x1, t2
+   LD a2,  AO1,  13 * SIZE
+   MADD3  t3, a4, x2, t3
+   LD y4,  YY,  15 * SIZE
+   MADD4  t4, a4, x1, t4
+   LD a4,  AO1,  15 * SIZE
+   ST t5,  YY,   4 * SIZE
+   ST t6,  YY,   5 * SIZE
+   ST t7,  YY,   6 * SIZE
+   ST t8,  YY,   7 * SIZE
+   addi.d  I, I, -1
+   addi.d  YY,  YY,   8 * SIZE
+   addi.d AO1, AO1,  8 * SIZE
+   blt $r0,    I, .L22
+   .align 3
+
+.L23:
+   ST t1,  YY,   0 * SIZE
+   MADD1  t1, a1, x1, y1
+   ST t2,  YY,   1 * SIZE
+   MADD2  t2, a1, x2, y2
+   ST t3,  YY,   2 * SIZE
+   MADD1  t3, a3, x1, y3
+   ST t4,  YY,   3 * SIZE
+   MADD2  t4, a3, x2, y4
+   MADD3  t1, a2, x2, t1
+   addi.d  AO1, AO1,  8 * SIZE
+   MADD4  t2, a2, x1, t2
+   addi.d  YY,  YY,   8 * SIZE
+   MADD3  t3, a4, x2, t3
+   MADD4  t4, a4, x1, t4
+   ST t1,  YY,  -4 * SIZE
+   ST t2,  YY,  -3 * SIZE
+   ST t3,  YY,  -2 * SIZE
+   ST t4,  YY,  -1 * SIZE
+   .align 3
+
+.L25:
+   andi    I,  M, 2
+   bge $r0,    I, .L26
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   LD y4,  YY,  3 * SIZE
+   MADD1  t1, a1, x1, y1
+   MADD2  t2, a1, x2, y2
+   MADD1  t3, a3, x1, y3
+   MADD2  t4, a3, x2, y4
+   MADD3  t1, a2, x2, t1
+   addi.d  YY,  YY,   4 * SIZE
+   MADD4  t2, a2, x1, t2
+   addi.d  AO1, AO1,  4 * SIZE
+   MADD3  t3, a4, x2, t3
+   MADD4  t4, a4, x1, t4
+   ST t1,  YY,  -4 * SIZE
+   ST t2,  YY,  -3 * SIZE
+   ST t3,  YY,  -2 * SIZE
+   ST t4,  YY,  -1 * SIZE
+   .align 3
+
+.L26:
+   andi    I,  M, 1
+   bge $r0,    I, .L900
+   LD y1,  YY,  0 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   MADD1  t1, a1, x1, y1
+   MADD2  t2, a1, x2, y2
+   MADD3  t1, a2, x2, t1
+   MADD4  t2, a2, x1, t2
+   ST t1,  YY,   0 * SIZE
+   ST t2,  YY,   1 * SIZE
+   .align 3
+
+.L900:
+   li  YORIG, 2 * SIZE
+   srai.d I,  M, 2
+   beq INCY, YORIG, .L999
+   move   XX, BUFFER
+   bge $r0,    I, .L905
+   .align 3
+
+.L902:
+   LD a1,  XX,  0 * SIZE
+   LD a2,  XX,  1 * SIZE
+   LD a3,  XX,  2 * SIZE
+   LD a4,  XX,  3 * SIZE
+   LD a5,  XX,  4 * SIZE
+   LD a6,  XX,  5 * SIZE
+   LD a7,  XX,  6 * SIZE
+   LD a8,  XX,  7 * SIZE
+   addi.d  I, I, -1
+   ST a1,  Y,  0 * SIZE
+   ST a2,  Y,  1 * SIZE
+   add.d   Y, Y, INCY
+   ST a3,  Y,  0 * SIZE
+   ST a4,  Y,  1 * SIZE
+   add.d   Y, Y, INCY
+   ST a5,  Y,  0 * SIZE
+   ST a6,  Y,  1 * SIZE
+   add.d   Y, Y, INCY
+   ST a7,  Y,  0 * SIZE
+   ST a8,  Y,  1 * SIZE
+   add.d   Y, Y, INCY
+   addi.d XX, XX, 8 * SIZE
+   blt $r0,    I, .L902
+   .align 3
+
+.L905:
+   andi    I,  M, 3
+   bge $r0,    I, .L999
+   .align 3
+
+.L906:
+   LD a1,  XX,  0 * SIZE
+   LD a2,  XX,  1 * SIZE
+   addi.d  XX, XX, 2 * SIZE
+   addi.d  I, I, -1
+   ST a1,  Y,  0 * SIZE
+   ST a2,  Y,  1 * SIZE
+   add.d  Y, Y, INCY
+   blt $r0,    I, .L906
+   .align 3
+
+.L999:
+   LDARG  $r23,  $sp,    0
+   LDARG  $r24,  $sp,    8
+   fld.d  $f24,  $sp,  16
+   fld.d  $f25,  $sp,  24
+#ifndef __64BIT__
+   fld.d  $f18,  $sp,  32
+   fld.d  $f19,  $sp,  40
+   fld.d  $f20,  $sp,  48
+   fld.d  $f21,  $sp,  56
+#endif
+#ifdef __64BIT__
+   addi.d  $sp, $sp, 32
+#else
+   addi.d  $sp, $sp, 64
+#endif
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/zgemv_t.S

@@ -0,0 +1,556 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define M          $r4
+#define N          $r5
+#define A          $r7
+#define LDA        $r8
+#define X          $r9
+#define INCX       $r10
+#define Y          $r11
+#define INCY       $r6
+#define BUFFER     $r17
+
+#define XORIG      $r18
+#define XX         $r12
+#define YY         $r13
+#define I          $r14
+#define J          $r15
+#define AO1        $r23
+#define AO2        $r24
+
+#define ALPHA_R    $f0
+#define ALPHA_I    $f1
+#define a1         $f22
+#define a2         $f8
+#define a3         $f23
+#define a4         $f9
+#define a5         $f10
+#define a6         $f11
+#define a7         $f12
+#define a8         $f13
+#define y1         $f14
+#define y2         $f15
+#define y3         $f16
+#define y4         $f17
+#define x1         $f3
+#define x2         $f4
+#define x3         $f2
+#define x4         $f5
+#define x5         $f6
+#define x6         $f7
+#define x7         $f18
+#define x8         $f19
+
+#if !defined(CONJ) && !defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        MADD
+#define    MADD3        NMSUB
+#define    MADD4        MADD
+#endif
+#if  defined(CONJ) && !defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        MADD
+#define    MADD3        MADD
+#define    MADD4        NMSUB
+#endif
+#if  !defined(CONJ) && defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        NMSUB
+#define    MADD3        MADD
+#define    MADD4        MADD
+#endif
+#if  defined(CONJ) && defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        NMSUB
+#define    MADD3        NMSUB
+#define    MADD4        NMSUB
+#endif
+
+   PROLOGUE
+
+   LDARG  INCY,    $sp,  0
+   LDARG  BUFFER,  $sp,  8
+#ifdef __64BIT__
+   addi.d  $sp, $sp, -16
+#else
+   addi.d  $sp, $sp, -32
+#endif
+   MTC  y1, $r0
+   SDARG  $r23,  $sp,    0
+   SDARG  $r24,  $sp,    8
+   slli.d     LDA,     LDA,  ZBASE_SHIFT
+#ifndef __64BIT__
+   fst.d  $f18,  $sp,  16
+   fst.d  $f19,  $sp,  24
+#endif
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    M, .L999
+   slli.d INCY, INCY, ZBASE_SHIFT
+   bge $r0,    N, .L999
+   li  I, 2 * SIZE
+   move   XORIG, X
+   beq INCX, I, .L10
+   srai.d  I,  M, 2
+   move    XORIG, BUFFER
+   move   YY, BUFFER
+   bge $r0,    I, .L05
+   .align 3
+
+.L02:
+   LD a1,  X,  0 * SIZE
+   LD a2,  X,  1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,  0 * SIZE
+   LD a4,  X,  1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,  0 * SIZE
+   LD a6,  X,  1 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,  0 * SIZE
+   LD a8,  X,  1 * SIZE
+   add.d   X, X, INCX
+   addi.d  I, I, -1
+   addi.d  YY, YY, 8 * SIZE
+   ST a1,  YY,  -8 * SIZE
+   ST a2,  YY,  -7 * SIZE
+   ST a3,  YY,  -6 * SIZE
+   ST a4,  YY,  -5 * SIZE
+   ST a5,  YY,  -4 * SIZE
+   ST a6,  YY,  -3 * SIZE
+   ST a7,  YY,  -2 * SIZE
+   ST a8,  YY,  -1 * SIZE
+   blt $r0,    I, .L02
+   .align 3
+
+.L05:
+   andi    I,  M, 3
+   bge $r0,    I, .L10
+   .align 3
+
+.L06:
+   LD a1,  X,  0 * SIZE
+   LD a2,  X,  1 * SIZE
+   add.d   X, X, INCX
+   ST a1,  YY,  0 * SIZE
+   ST a2,  YY,  1 * SIZE
+   addi.d  I, I, -1
+   addi.d YY, YY, 2 * SIZE
+   blt $r0,    I, .L06
+   .align 3
+
+.L10:
+   srai.d  J,  N, 1
+   move   YY, Y
+   bge $r0,    J, .L20
+   .align 3
+
+.L11:
+   move    AO1, A
+   MOV y2, y1
+   add.d   AO2, A,      LDA
+   MOV y3, y1
+   add.d   A,   AO2,    LDA
+   MOV y4, y1
+   srai.d  I,  M, 2
+   move   XX, XORIG
+   bge $r0,    I, .L15
+   LD x1,  XX,  0 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD x4,  XX,  3 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD a3,  AO2,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD a4,  AO2,  1 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   LD a7,  AO2,  2 * SIZE
+   LD a6,  AO1,  3 * SIZE
+   LD a8,  AO2,  3 * SIZE
+   addi.d  I, I, -1
+   bge $r0,    I, .L13
+   .align  3
+.L12:
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,  2 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a1,  AO1,  4 * SIZE
+   MADD1  y3, a3, x1, y3
+   MADD2  y4, a3, x2, y4
+   LD a3,  AO2,  4 * SIZE
+   MADD3  y1, a2, x2, y1
+   MADD4  y2, a2, x1, y2
+   LD a2,  AO1,  5 * SIZE
+   MADD3  y3, a4, x2, y3
+   LD x2,  XX,  5 * SIZE
+   MADD4  y4, a4, x1, y4
+   LD a4,  AO2,  5 * SIZE
+   MADD1  y1, a5, x3, y1
+   LD x1,  XX,  4 * SIZE
+   MADD2  y2, a5, x4, y2
+   LD a5,  AO1,  6 * SIZE
+   MADD1  y3, a7, x3, y3
+   MADD2  y4, a7, x4, y4
+   LD a7,  AO2,  6 * SIZE
+   MADD3  y1, a6, x4, y1
+   addi.d  I, I, -1
+   MADD4  y2, a6, x3, y2
+   LD a6,  AO1,  7 * SIZE
+   MADD3  y3, a8, x4, y3
+   LD x4,  XX,  7 * SIZE
+   MADD4  y4, a8, x3, y4
+   LD a8,  AO2,  7 * SIZE
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,  6 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a1,  AO1,   8 * SIZE
+   MADD1  y3, a3, x1, y3
+   MADD2  y4, a3, x2, y4
+   LD a3,  AO2,   8 * SIZE
+   MADD3  y1, a2, x2, y1
+   MADD4  y2, a2, x1, y2
+   LD a2,  AO1,   9 * SIZE
+   MADD3  y3, a4, x2, y3
+   LD x2,  XX,   9 * SIZE
+   MADD4  y4, a4, x1, y4
+   LD a4,  AO2,   9 * SIZE
+   MADD1  y1, a5, x3, y1
+   LD x1,  XX,   8 * SIZE
+   MADD2  y2, a5, x4, y2
+   LD a5,  AO1,  10 * SIZE
+   MADD1  y3, a7, x3, y3
+   addi.d  XX,  XX,   8 * SIZE
+   MADD2  y4, a7, x4, y4
+   LD a7,  AO2,  10 * SIZE
+   MADD3  y1, a6, x4, y1
+   addi.d  AO2, AO2,  8 * SIZE
+   MADD4  y2, a6, x3, y2
+   LD a6,  AO1,  11 * SIZE
+   MADD3  y3, a8, x4, y3
+   LD x4,  XX,   3 * SIZE
+   MADD4  y4, a8, x3, y4
+   LD a8,  AO2,   3 * SIZE
+   addi.d AO1, AO1,  8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,   2 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a1,  AO1,  4 * SIZE
+   MADD1  y3, a3, x1, y3
+   MADD2  y4, a3, x2, y4
+   LD a3,  AO2,  4 * SIZE
+   MADD3  y1, a2, x2, y1
+   MADD4  y2, a2, x1, y2
+   LD a2,  AO1,  5 * SIZE
+   MADD3  y3, a4, x2, y3
+   LD x2,  XX,  5 * SIZE
+   MADD4  y4, a4, x1, y4
+   LD a4,  AO2,  5 * SIZE
+   MADD1  y1, a5, x3, y1
+   LD x1,  XX,  4 * SIZE
+   MADD2  y2, a5, x4, y2
+   LD a5,  AO1,  6 * SIZE
+   MADD1  y3, a7, x3, y3
+   MADD2  y4, a7, x4, y4
+   LD a7,  AO2,  6 * SIZE
+   MADD3  y1, a6, x4, y1
+   MADD4  y2, a6, x3, y2
+   LD a6,  AO1,  7 * SIZE
+   MADD3  y3, a8, x4, y3
+   LD x4,  XX,  7 * SIZE
+   MADD4  y4, a8, x3, y4
+   LD a8,  AO2,  7 * SIZE
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,  6 * SIZE
+   MADD2  y2, a1, x2, y2
+   MADD1  y3, a3, x1, y3
+   MADD2  y4, a3, x2, y4
+   MADD3  y1, a2, x2, y1
+   MADD4  y2, a2, x1, y2
+   MADD3  y3, a4, x2, y3
+   MADD4  y4, a4, x1, y4
+   MADD1  y1, a5, x3, y1
+   MADD2  y2, a5, x4, y2
+   MADD1  y3, a7, x3, y3
+   MADD2  y4, a7, x4, y4
+   MADD3  y1, a6, x4, y1
+   addi.d  XX,  XX,   8 * SIZE
+   MADD4  y2, a6, x3, y2
+   addi.d  AO1, AO1,  8 * SIZE
+   MADD3  y3, a8, x4, y3
+   addi.d  AO2, AO2,  8 * SIZE
+   MADD4  y4, a8, x3, y4
+   .align 3
+
+.L15:
+   andi    I,  M, 2
+   bge $r0,    I, .L17
+   LD x1,  XX,  0 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD x3,  XX,  2 * SIZE
+   LD x4,  XX,  3 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD a3,  AO2,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD a4,  AO2,  1 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   LD a7,  AO2,  2 * SIZE
+   LD a6,  AO1,  3 * SIZE
+   LD a8,  AO2,  3 * SIZE
+   MADD1  y1, a1, x1, y1
+   MADD2  y2, a1, x2, y2
+   MADD1  y3, a3, x1, y3
+   MADD2  y4, a3, x2, y4
+   MADD3  y1, a2, x2, y1
+   MADD4  y2, a2, x1, y2
+   MADD3  y3, a4, x2, y3
+   MADD4  y4, a4, x1, y4
+   MADD1  y1, a5, x3, y1
+   MADD2  y2, a5, x4, y2
+   MADD1  y3, a7, x3, y3
+   MADD2  y4, a7, x4, y4
+   MADD3  y1, a6, x4, y1
+   addi.d  XX,  XX,   4 * SIZE
+   MADD4  y2, a6, x3, y2
+   addi.d  AO1, AO1,  4 * SIZE
+   MADD3  y3, a8, x4, y3
+   addi.d  AO2, AO2,  4 * SIZE
+   MADD4  y4, a8, x3, y4
+   .align 3
+
+.L17:
+   andi    I,  M, 1
+.align 3
+
+   bge $r0,    I, .L19
+.L18:
+   LD x1,  XX,  0 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD a3,  AO2,  0 * SIZE
+   MADD1  y1, a1, x1, y1
+   LD a2,  AO1,  1 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a4,  AO2,  1 * SIZE
+   MADD1  y3, a3, x1, y3
+   MADD2  y4, a3, x2, y4
+   MADD3  y1, a2, x2, y1
+   MADD4  y2, a2, x1, y2
+   MADD3  y3, a4, x2, y3
+   MADD4  y4, a4, x1, y4
+   .align 3
+
+.L19:
+   LD a1,  Y,  0 * SIZE
+   LD a2,  Y,  1 * SIZE
+   add.d   Y, Y, INCY
+   LD a3,  Y,  0 * SIZE
+   LD a4,  Y,  1 * SIZE
+   add.d   Y, Y, INCY
+   MADD  a1, y1, ALPHA_R, a1
+   MADD  a2, y1, ALPHA_I, a2
+   MADD  a3, y3, ALPHA_R, a3
+   MADD  a4, y3, ALPHA_I, a4
+   NMSUB  a1, y2, ALPHA_I, a1
+   MADD  a2, y2, ALPHA_R, a2
+   NMSUB  a3, y4, ALPHA_I, a3
+   MTC  y1, $r0
+   MADD  a4, y4, ALPHA_R, a4
+   addi.d  J, J, -1
+   ST a1,  YY,   0 * SIZE
+   ST a2,  YY,   1 * SIZE
+   add.d   YY, YY, INCY
+   ST a3,  YY,   0 * SIZE
+   ST a4,  YY,   1 * SIZE
+   add.d  YY, YY, INCY
+   blt $r0,    J, .L11
+   .align 3
+
+.L20:
+   andi    J,  N, 1
+   MOV y2, y1
+   srai.d I,  M, 2
+   bge $r0,    J, .L999
+   MOV y3, y1
+   move    AO1, A
+   MOV y4, y1
+   move   XX, XORIG
+   bge $r0,    I, .L25
+   LD a1,  AO1,  0 * SIZE
+   LD x1,  XX,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   LD x4,  XX,  3 * SIZE
+   addi.d  I, I, -1
+   LD a6,  AO1,  3 * SIZE
+   bge $r0,    I, .L23
+   .align  3
+.L22:
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,  2 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a1,  AO1,  4 * SIZE
+   MADD3  y3, a2, x2, y3
+   LD x2,  XX,  5 * SIZE
+   MADD4  y4, a2, x1, y4
+   LD a2,  AO1,  5 * SIZE
+   MADD1  y1, a5, x3, y1
+   LD x1,  XX,  4 * SIZE
+   MADD2  y2, a5, x4, y2
+   LD a5,  AO1,  6 * SIZE
+   MADD3  y3, a6, x4, y3
+   LD x4,  XX,  7 * SIZE
+   MADD4  y4, a6, x3, y4
+   LD a6,  AO1,  7 * SIZE
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,  6 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a1,  AO1,   8 * SIZE
+   MADD3  y3, a2, x2, y3
+   LD x2,  XX,   9 * SIZE
+   MADD4  y4, a2, x1, y4
+   LD a2,  AO1,   9 * SIZE
+   MADD1  y1, a5, x3, y1
+   LD x1,  XX,   8 * SIZE
+   MADD2  y2, a5, x4, y2
+   LD a5,  AO1,  10 * SIZE
+   MADD3  y3, a6, x4, y3
+   LD x4,  XX,  11 * SIZE
+   MADD4  y4, a6, x3, y4
+   LD a6,  AO1,  11 * SIZE
+   addi.d  I, I, -1
+   addi.d  XX,  XX,   8 * SIZE
+   addi.d AO1, AO1,  8 * SIZE
+   blt $r0,    I, .L22
+   .align 3
+
+.L23:
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,   2 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a1,  AO1,  4 * SIZE
+   MADD3  y3, a2, x2, y3
+   LD x2,  XX,  5 * SIZE
+   MADD4  y4, a2, x1, y4
+   LD a2,  AO1,  5 * SIZE
+   MADD1  y1, a5, x3, y1
+   LD x1,  XX,  4 * SIZE
+   MADD2  y2, a5, x4, y2
+   LD a5,  AO1,  6 * SIZE
+   MADD3  y3, a6, x4, y3
+   LD x4,  XX,  7 * SIZE
+   MADD4  y4, a6, x3, y4
+   LD a6,  AO1,  7 * SIZE
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,  6 * SIZE
+   MADD2  y2, a1, x2, y2
+   MADD3  y3, a2, x2, y3
+   MADD4  y4, a2, x1, y4
+   MADD1  y1, a5, x3, y1
+   MADD2  y2, a5, x4, y2
+   MADD3  y3, a6, x4, y3
+   addi.d  XX,  XX,   8 * SIZE
+   MADD4  y4, a6, x3, y4
+   addi.d  AO1, AO1,  8 * SIZE
+   .align 3
+
+.L25:
+   andi    I,  M, 2
+   bge $r0,    I, .L27
+   LD a1,  AO1,  0 * SIZE
+   LD x1,  XX,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,  2 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a6,  AO1,  3 * SIZE
+   MADD3  y3, a2, x2, y3
+   LD x4,  XX,  3 * SIZE
+   MADD4  y4, a2, x1, y4
+   MADD1  y1, a5, x3, y1
+   MADD2  y2, a5, x4, y2
+   MADD3  y3, a6, x4, y3
+   addi.d  XX,  XX,   4 * SIZE
+   MADD4  y4, a6, x3, y4
+   addi.d  AO1, AO1,  4 * SIZE
+   .align 3
+
+.L27:
+   andi    I,  M, 1
+.align 3
+
+   bge $r0,    I, .L29
+.L28:
+   LD a1,  AO1,  0 * SIZE
+   LD x1,  XX,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD x2,  XX,  1 * SIZE
+   MADD1  y1, a1, x1, y1
+   MADD2  y2, a1, x2, y2
+   MADD3  y3, a2, x2, y3
+   MADD4  y4, a2, x1, y4
+   .align 3
+
+.L29:
+   LD a1,  Y,  0 * SIZE
+   LD a2,  Y,  1 * SIZE
+   ADD y1, y1, y3
+   ADD y2, y2, y4
+   MADD  a1, y1, ALPHA_R, a1
+   MADD  a2, y1, ALPHA_I, a2
+   NMSUB  a1, y2, ALPHA_I, a1
+   MADD  a2, y2, ALPHA_R, a2
+   ST a1,  YY,   0 * SIZE
+   ST a2,  YY,   1 * SIZE
+   .align 3
+
+.L999:
+   LDARG  $r23,  $sp,    0
+   LDARG  $r24,  $sp,    8
+#ifndef __64BIT__
+   fld.d  $f18,  $sp,  16
+   fld.d  $f19,  $sp,  24
+#endif
+#ifdef __64BIT__
+   addi.d  $sp, $sp, 16
+#else
+   addi.d  $sp, $sp, 32
+#endif
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/znrm2.S

@@ -0,0 +1,304 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define XX     $r7
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+#define ALPHA  $f4
+#define max    $f5
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    INCX, .L999
+   move    XX, X
+   MOV s2, s1
+   srai.d  I, N, 2
+   MOV s3, s1
+   MOV    s4, s1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   LD a8,  X,   1 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   NOP
+   FABS    t3, a3
+   LD a2,  X,   1 * SIZE
+   FABS    t4, a4
+   add.d   X, X, INCX
+   CMPLT   $fcc0, s1, t1
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t2
+   NOP
+   CMPLT   $fcc2, s3, t3
+   LD a4,  X,   1 * SIZE
+   CMPLT   $fcc3, s4, t4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   FABS    t1, a5
+   LD a5,  X,   0 * SIZE
+   FABS    t2, a6
+   NOP
+   FABS    t3, a7
+   LD a6,  X,   1 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+   CMPLT   $fcc0, s1, t1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t2
+   NOP
+   CMPLT   $fcc2, s3, t3
+   LD a8,  X,   1 * SIZE
+   CMPLT   $fcc3, s4, t4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   addi.d  I, I, -1
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMPLT   $fcc2, s3, t3
+   CMPLT   $fcc3, s4, t4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   FABS    t1, a5
+   FABS    t2, a6
+   FABS    t3, a7
+   FABS    t4, a8
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMPLT   $fcc2, s3, t3
+   CMPLT   $fcc3, s4, t4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   .align 3
+
+.L15:
+   andi    I,  N, 3
+   bge $r0,    I, .L100
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   addi.d  I, I, -1
+   FABS    t1, a1
+   FABS    t2, a2
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+
+.L100:
+   CMPLT   $fcc0, s1, s2
+   CMPLT   $fcc1, s3, s4
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+   CMPLT   $fcc0, s1, s3
+   CMOVT  s1,  s1,  s3,  $fcc0
+   lu12i.w TEMP, 0x3f800
+   movgr2fr.d  a1,   $r0
+   movgr2fr.w  ALPHA, TEMP
+   CMPEQ   $fcc0, s1, a1
+   fcvt.d.s   ALPHA, ALPHA
+   bcnez   $fcc0, .L999
+   fdiv.d  ALPHA, ALPHA, s1
+   MOV max, s1
+   MOV s1, a1
+   MOV s2, a1
+   MOV s3, a1
+   MOV s4, a1
+   srai.d  I, N, 2
+   bge $r0,    I, .L105
+   LD a1,  XX,   0 * SIZE
+   LD a2,  XX,   1 * SIZE
+   add.d   XX, XX, INCX
+   LD a3,  XX,   0 * SIZE
+   LD a4,  XX,   1 * SIZE
+   add.d   XX, XX, INCX
+   LD a5,  XX,   0 * SIZE
+   LD a6,  XX,   1 * SIZE
+   add.d   XX, XX, INCX
+   LD a7,  XX,   0 * SIZE
+   LD a8,  XX,   1 * SIZE
+   addi.d  I, I, -1
+   add.d  XX, XX, INCX
+   bge $r0,    I, .L104
+   .align 3
+
+.L103:
+   MUL t1, ALPHA, a1
+   LD a1,  XX,   0 * SIZE
+   MUL t2, ALPHA, a2
+   addi.d  I, I, -1
+   MUL t3, ALPHA, a3
+   LD a2,  XX,   1 * SIZE
+   MUL t4, ALPHA, a4
+   add.d   XX, XX, INCX
+   MADD  s1, t1, t1, s1
+   LD a3,  XX,   0 * SIZE
+   MADD  s2, t2, t2, s2
+   NOP
+   MADD  s3, t3, t3, s3
+   LD a4,  XX,   1 * SIZE
+   MADD  s4, t4, t4, s4
+   add.d   XX, XX, INCX
+   MUL t1, ALPHA, a5
+   LD a5,  XX,   0 * SIZE
+   MUL t2, ALPHA, a6
+   NOP
+   MUL t3, ALPHA, a7
+   LD a6,  XX,   1 * SIZE
+   MUL t4, ALPHA, a8
+   add.d   XX, XX, INCX
+   MADD  s1, t1, t1, s1
+   LD a7,  XX,   0 * SIZE
+   MADD  s2, t2, t2, s2
+   LD a8,  XX,   1 * SIZE
+   MADD  s3, t3, t3, s3
+   add.d   XX, XX, INCX
+   MADD  s4, t4, t4, s4
+   blt $r0,    I, .L103
+   .align 3
+
+.L104:
+   MUL t1, ALPHA, a1
+   MUL t2, ALPHA, a2
+   MUL t3, ALPHA, a3
+   MUL t4, ALPHA, a4
+   MADD  s1, t1, t1, s1
+   MADD  s2, t2, t2, s2
+   MADD  s3, t3, t3, s3
+   MADD  s4, t4, t4, s4
+   MUL t1, ALPHA, a5
+   MUL t2, ALPHA, a6
+   MUL t3, ALPHA, a7
+   MUL t4, ALPHA, a8
+   MADD  s1, t1, t1, s1
+   MADD  s2, t2, t2, s2
+   MADD  s3, t3, t3, s3
+   MADD  s4, t4, t4, s4
+   .align 3
+
+.L105:
+   andi    I,  N, 3
+   bge $r0,    I, .L998
+   .align  3
+
+.L106:
+   LD a1,  XX,   0 * SIZE
+   LD a2,  XX,   1 * SIZE
+   addi.d  I, I, -1
+   MUL t1, ALPHA, a1
+   MUL t2, ALPHA, a2
+   MADD  s1, t1, t1, s1
+   add.d   XX, XX, INCX
+   MADD  s2, t2, t2, s2
+   blt $r0,    I, .L106
+   .align 3
+
+.L998:
+   ADD s1, s1, s2
+   ADD s3, s3, s4
+   ADD s1, s1, s3
+   fsqrt.d s1, s1
+   move $r4, $r17
+   MUL $f0, max, s1
+   jirl    $r0, $r1, 0x0
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/loongarch64/zscal.S

@@ -0,0 +1,356 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N          $r4
+#define X          $r7
+#define INCX       $r8
+#define I          $r17
+#define TEMP       $r18
+#define XX         $r5
+#define ALPHA_R    $f0
+#define ALPHA_I    $f1
+#define a1         $f22
+#define a2         $f8
+#define a3         $f23
+#define a4         $f9
+#define a5         $f10
+#define a6         $f11
+#define a7         $f12
+#define a8         $f13
+#define t1         $f14
+#define t2         $f15
+#define t3         $f16
+#define t4         $f17
+
+   PROLOGUE
+
+   li  TEMP, 2 * SIZE
+   MTC  a1, $r0
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    N, .L999
+   CMPEQ   $fcc0, ALPHA_R, a1
+   CMPEQ   $fcc1, ALPHA_I, a1
+   bceqz   $fcc0, .L50
+   bceqz   $fcc1, .L50
+   srai.d I, N, 2
+   bne INCX, TEMP, .L20
+   bge $r0,    I, .L15
+   .align 3
+
+.L12:
+   ST a1,  X,   0 * SIZE
+   ST a1,  X,   1 * SIZE
+   ST a1,  X,   2 * SIZE
+   ST a1,  X,   3 * SIZE
+   ST a1,  X,   4 * SIZE
+   ST a1,  X,   5 * SIZE
+   ST a1,  X,   6 * SIZE
+   ST a1,  X,   7 * SIZE
+   addi.w  I, I, -1
+   addi.d X, X, 8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L15:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   .align  3
+.L16:
+   ST a1,  X,   0 * SIZE
+   ST a1,  X,   1 * SIZE
+   addi.d  I, I, -1
+   addi.d X, X, 2 * SIZE
+   blt $r0,    I, .L16
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   .align 3
+
+.L20:
+   srai.d  I, N, 2
+   bge $r0,    I, .L25
+   .align 3
+
+.L22:
+   ST a1,  X,   0 * SIZE
+   ST a1,  X,   1 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   ST a1,  X,   1 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   ST a1,  X,   1 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   ST a1,  X,   1 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L22
+   .align 3
+
+.L25:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   .align  3
+.L26:
+   ST a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   ST a1,  X,   1 * SIZE
+   add.d  X, X, INCX
+   blt $r0,    I, .L26
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   .align 3
+
+.L50:
+   srai.d I, N, 2
+   bne INCX, TEMP, .L60
+   addi.d I, I, -1
+   blt I,  $r0, .L55
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   LD a5,  X,   4 * SIZE
+   LD a6,  X,   5 * SIZE
+   MUL t1, ALPHA_R, a1
+   LD a7,  X,   6 * SIZE
+   MUL t2, ALPHA_I, a1
+   LD a8,  X,   7 * SIZE
+   MUL t3, ALPHA_R, a3
+   MUL    t4, ALPHA_I, a3
+   bge $r0,    I, .L53
+   .align 3
+
+.L52:
+   NMSUB  t1, a2, ALPHA_I, t1
+   LD a1,  X,   8 * SIZE
+   MADD  t2, a2, ALPHA_R, t2
+   LD a2,  X,   9 * SIZE
+   NMSUB  t3, a4, ALPHA_I, t3
+   LD a3,  X,  10 * SIZE
+   MADD  t4, a4, ALPHA_R, t4
+   LD a4,  X,  11 * SIZE
+   ST t1,  X,   0 * SIZE
+   MUL t1, ALPHA_R, a5
+   ST t2,  X,   1 * SIZE
+   MUL t2, ALPHA_I, a5
+   ST t3,  X,   2 * SIZE
+   MUL t3, ALPHA_R, a7
+   ST t4,  X,   3 * SIZE
+   MUL t4, ALPHA_I, a7
+   NMSUB  t1, a6, ALPHA_I, t1
+   LD a5,  X,  12 * SIZE
+   MADD  t2, a6, ALPHA_R, t2
+   LD a6,  X,  13 * SIZE
+   NMSUB  t3, a8, ALPHA_I, t3
+   LD a7,  X,  14 * SIZE
+   MADD  t4, a8, ALPHA_R, t4
+   LD a8,  X,  15 * SIZE
+   ST t1,  X,   4 * SIZE
+   MUL t1, ALPHA_R, a1
+   ST t2,  X,   5 * SIZE
+   MUL t2, ALPHA_I, a1
+   ST t3,  X,   6 * SIZE
+   MUL t3, ALPHA_R, a3
+   ST t4,  X,   7 * SIZE
+   MUL t4, ALPHA_I, a3
+   addi.d  I, I, -1
+   addi.d X, X, 8 * SIZE
+   blt $r0,    I, .L52
+   .align 3
+
+.L53:
+   NMSUB  t1, a2, ALPHA_I, t1
+   MADD  t2, a2, ALPHA_R, t2
+   NMSUB  t3, a4, ALPHA_I, t3
+   MADD  t4, a4, ALPHA_R, t4
+   ST t1,  X,   0 * SIZE
+   MUL t1, ALPHA_R, a5
+   ST t2,  X,   1 * SIZE
+   MUL t2, ALPHA_I, a5
+   ST t3,  X,   2 * SIZE
+   MUL t3, ALPHA_R, a7
+   ST t4,  X,   3 * SIZE
+   MUL t4, ALPHA_I, a7
+   NMSUB  t1, a6, ALPHA_I, t1
+   MADD  t2, a6, ALPHA_R, t2
+   NMSUB  t3, a8, ALPHA_I, t3
+   MADD  t4, a8, ALPHA_R, t4
+   ST t1,  X,   4 * SIZE
+   ST t2,  X,   5 * SIZE
+   ST t3,  X,   6 * SIZE
+   ST t4,  X,   7 * SIZE
+   addi.d  X, X, 8 * SIZE
+   .align 3
+
+.L55:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   .align  3
+.L56:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   MUL t1, ALPHA_R, a1
+   MUL t2, ALPHA_I, a1
+   NMSUB  t1, a2, ALPHA_I, t1
+   MADD  t2, a2, ALPHA_R, t2
+   addi.d  X, X, 2 * SIZE
+   addi.d  I, I, -1
+   ST t1,  X,  -2 * SIZE
+   ST t2,  X,  -1 * SIZE
+   blt $r0,    I, .L56
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   .align 3
+
+.L60:
+   srai.d  I, N, 2
+   move    XX, X
+   addi.d I, I, -1
+   blt I,  $r0, .L65
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   MUL t1, ALPHA_R, a1
+   LD a7,  X,   0 * SIZE
+   MUL t2, ALPHA_I, a1
+   LD a8,  X,   1 * SIZE
+   MUL t3, ALPHA_R, a3
+   add.d   X, X, INCX
+   MUL    t4, ALPHA_I, a3
+   bge $r0,    I, .L63
+   .align 3
+
+.L62:
+   NMSUB  t1, a2, ALPHA_I, t1
+   LD a1,  X,   0 * SIZE
+   MADD  t2, a2, ALPHA_R, t2
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   NMSUB  t3, a4, ALPHA_I, t3
+   LD a3,  X,   0 * SIZE
+   MADD  t4, a4, ALPHA_R, t4
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   ST t1,  XX,   0 * SIZE
+   MUL t1, ALPHA_R, a5
+   ST t2,  XX,   1 * SIZE
+   MUL t2, ALPHA_I, a5
+   add.d   XX, XX, INCX
+   ST t3,  XX,   0 * SIZE
+   MUL t3, ALPHA_R, a7
+   ST t4,  XX,   1 * SIZE
+   MUL t4, ALPHA_I, a7
+   add.d   XX, XX, INCX
+   NMSUB  t1, a6, ALPHA_I, t1
+   LD a5,  X,   0 * SIZE
+   MADD  t2, a6, ALPHA_R, t2
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   NMSUB  t3, a8, ALPHA_I, t3
+   LD a7,  X,   0 * SIZE
+   MADD  t4, a8, ALPHA_R, t4
+   LD a8,  X,   1 * SIZE
+   add.d   X, X, INCX
+   ST t1,  XX,   0 * SIZE
+   MUL t1, ALPHA_R, a1
+   ST t2,  XX,   1 * SIZE
+   MUL t2, ALPHA_I, a1
+   add.d   XX, XX, INCX
+   ST t3,  XX,   0 * SIZE
+   MUL t3, ALPHA_R, a3
+   ST t4,  XX,   1 * SIZE
+   MUL t4, ALPHA_I, a3
+   addi.d  I, I, -1
+   add.d  XX, XX, INCX
+   blt $r0,    I, .L62
+   .align 3
+
+.L63:
+   NMSUB  t1, a2, ALPHA_I, t1
+   MADD  t2, a2, ALPHA_R, t2
+   NMSUB  t3, a4, ALPHA_I, t3
+   MADD  t4, a4, ALPHA_R, t4
+   ST t1,  XX,   0 * SIZE
+   MUL t1, ALPHA_R, a5
+   ST t2,  XX,   1 * SIZE
+   MUL t2, ALPHA_I, a5
+   add.d   XX, XX, INCX
+   ST t3,  XX,   0 * SIZE
+   MUL t3, ALPHA_R, a7
+   ST t4,  XX,   1 * SIZE
+   MUL t4, ALPHA_I, a7
+   add.d   XX, XX, INCX
+   NMSUB  t1, a6, ALPHA_I, t1
+   MADD  t2, a6, ALPHA_R, t2
+   NMSUB  t3, a8, ALPHA_I, t3
+   MADD  t4, a8, ALPHA_R, t4
+   ST t1,  XX,   0 * SIZE
+   ST t2,  XX,   1 * SIZE
+   add.d   XX, XX, INCX
+   ST t3,  XX,   0 * SIZE
+   ST t4,  XX,   1 * SIZE
+   add.d   XX, XX, INCX
+   .align 3
+
+.L65:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   .align  3
+.L66:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   MUL t1, ALPHA_R, a1
+   MUL t2, ALPHA_I, a1
+   NMSUB  t1, a2, ALPHA_I, t1
+   MADD  t2, a2, ALPHA_R, t2
+   addi.d  I, I, -1
+   ST t1,  X,   0 * SIZE
+   ST t2,  X,   1 * SIZE
+   add.d  X, X, INCX
+   blt $r0,    I, .L66
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE

kernel/x86_64/sgemv_t_4.c

@@ -38,7 +38,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "sgemv_t_microk_haswell-4.c"
 #elif defined (SKYLAKEX) || defined (COOPERLAKE)
 #include "sgemv_t_microk_haswell-4.c"
-#include "sgemv_t_microk_skylakex.c"
+/*#include "sgemv_t_microk_skylakex.c"*/
 #endif
 
 #if defined(STEAMROLLER) || defined(EXCAVATOR)

lapack-netlib/TESTING/EIG/cchkee.F

@@ -1075,7 +1075,8 @@
       CHARACTER*80       LINE
       INTEGER            I, I1, IC, INFO, ITMP, K, LENP, MAXTYP, NEWSD,
      $                   NK, NN, NPARMS, NRHS, NTYPES,
-     $                   VERS_MAJOR, VERS_MINOR, VERS_PATCH, N_THREADS
+     $                   VERS_MAJOR, VERS_MINOR, VERS_PATCH
+      INTEGER*4          N_THREADS, ONE_THREAD
       REAL               EPS, S1, S2, THRESH, THRSHN
 *     ..
 *     .. Local Arrays ..
@@ -1872,7 +1873,8 @@
          IF( TSTERR ) THEN
 #if defined(_OPENMP)
             N_THREADS = OMP_GET_MAX_THREADS()
-            CALL OMP_SET_NUM_THREADS(1)
+            ONE_THREAD = 1
+            CALL OMP_SET_NUM_THREADS(ONE_THREAD)
 #endif
             CALL CERRST( 'CST', NOUT )
 #if defined(_OPENMP)
@@ -2339,7 +2341,8 @@
          IF( TSTERR ) THEN
 #if defined(_OPENMP)
             N_THREADS = OMP_GET_MAX_THREADS()
-            CALL OMP_SET_NUM_THREADS(1)
+            ONE_THREAD = 1
+            CALL OMP_SET_NUM_THREADS(ONE_THREAD)
 #endif
             CALL CERRST( 'CHB', NOUT )
 #if defined(_OPENMP)

lapack-netlib/TESTING/EIG/dchkee.F

@@ -1081,7 +1081,8 @@
       CHARACTER*80       LINE
       INTEGER            I, I1, IC, INFO, ITMP, K, LENP, MAXTYP, NEWSD,
      $                   NK, NN, NPARMS, NRHS, NTYPES,
-     $                   VERS_MAJOR, VERS_MINOR, VERS_PATCH, N_THREADS
+     $                   VERS_MAJOR, VERS_MINOR, VERS_PATCH
+      INTEGER*4          N_THREADS, ONE_THREAD
       DOUBLE PRECISION   EPS, S1, S2, THRESH, THRSHN
 *     ..
 *     .. Local Arrays ..
@@ -1877,7 +1878,8 @@
          IF( TSTERR ) THEN
 #if defined(_OPENMP)
             N_THREADS = OMP_GET_MAX_THREADS()
-            CALL OMP_SET_NUM_THREADS(1)
+            ONE_THREAD = 1
+            CALL OMP_SET_NUM_THREADS(ONE_THREAD)
 #endif
             CALL DERRST( 'DST', NOUT )
 #if defined(_OPENMP)

lapack-netlib/TESTING/EIG/schkee.F

@@ -1081,7 +1081,8 @@
       CHARACTER*80       LINE
       INTEGER            I, I1, IC, INFO, ITMP, K, LENP, MAXTYP, NEWSD,
      $                   NK, NN, NPARMS, NRHS, NTYPES,
-     $                   VERS_MAJOR, VERS_MINOR, VERS_PATCH, N_THREADS
+     $                   VERS_MAJOR, VERS_MINOR, VERS_PATCH
+      INTEGER*4          N_THREADS, ONE_THREAD
       REAL               EPS, S1, S2, THRESH, THRSHN
 *     ..
 *     .. Local Arrays ..
@@ -1878,7 +1879,8 @@
          IF( TSTERR ) THEN
 #if defined(_OPENMP)
             N_THREADS = OMP_GET_MAX_THREADS()
-            CALL OMP_SET_NUM_THREADS(1)
+            ONE_THREAD = 1
+            CALL OMP_SET_NUM_THREADS(ONE_THREAD)
 #endif
             CALL SERRST( 'SST', NOUT )
 #if defined(_OPENMP)

lapack-netlib/TESTING/EIG/zchkee.F

@@ -1075,7 +1075,8 @@
       CHARACTER*80       LINE
       INTEGER            I, I1, IC, INFO, ITMP, K, LENP, MAXTYP, NEWSD,
      $                   NK, NN, NPARMS, NRHS, NTYPES,
-     $                   VERS_MAJOR, VERS_MINOR, VERS_PATCH, N_THREADS
+     $                   VERS_MAJOR, VERS_MINOR, VERS_PATCH
+      INTEGER*4          N_THREADS, ONE_THREAD
       DOUBLE PRECISION   EPS, S1, S2, THRESH, THRSHN
 *     ..
 *     .. Local Arrays ..
@@ -1872,7 +1873,8 @@
          IF( TSTERR ) THEN
 #if defined(_OPENMP)
             N_THREADS = OMP_GET_MAX_THREADS()
-            CALL OMP_SET_NUM_THREADS(1)
+            ONE_THREAD = 1
+            CALL OMP_SET_NUM_THREADS(ONE_THREAD)
 #endif
             CALL ZERRST( 'ZST', NOUT )
 #if defined(_OPENMP)
@@ -2337,7 +2339,8 @@
          IF( TSTERR ) THEN
 #if defined(_OPENMP)
             N_THREADS = OMP_GET_MAX_THREADS()
-            CALL OMP_SET_NUM_THREADS(1)
+            ONE_THREAD = 1
+            CALL OMP_SET_NUM_THREADS(ONE_THREAD)
 #endif
             CALL ZERRST( 'ZHB', NOUT )
 #if defined(_OPENMP)

lapack/laswp/loongarch64/Makefile

@@ -0,0 +1,12 @@
+TOPDIR	= ../../..
+include ../../../Makefile.system
+
+ifndef LASWP
+LASWP	= ../generic/laswp_k.c
+endif
+
+ifndef ZLASWP
+ZLASWP	= ../generic/zlaswp_k.c
+endif
+
+include ../generic/Makefile

openblas_config_template.h

@@ -99,6 +99,8 @@ typedef int blasint;
 
 /* Inclusion of Linux-specific header is needed for definition of cpu_set_t. */
 #ifdef OPENBLAS_OS_LINUX
-#define _GNU_SOURCE
+#ifndef _GNU_SOURCE
+ #define _GNU_SOURCE
+#endif
 #include <sched.h>
 #endif

param.h

@@ -2502,7 +2502,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define GEMM_DEFAULT_OFFSET_A 0
 #define GEMM_DEFAULT_OFFSET_B 2048
-#define GEMM_DEFAULT_ALIGN (BLASLONG)0x03fffUL
+#define GEMM_DEFAULT_ALIGN 0x03fffUL
 
 #define SGEMM_DEFAULT_UNROLL_M 2
 #define SGEMM_DEFAULT_UNROLL_N 8
@@ -2534,7 +2534,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define GEMM_DEFAULT_OFFSET_A 0
 #define GEMM_DEFAULT_OFFSET_B 2048
-#define GEMM_DEFAULT_ALIGN (BLASLONG)0x03fffUL
+#define GEMM_DEFAULT_ALIGN 0x03fffUL
 
 #define SGEMM_DEFAULT_UNROLL_M 4
 #define SGEMM_DEFAULT_UNROLL_N 4
@@ -2691,6 +2691,52 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SYMV_P	16
 #endif
 
+#if defined (LOONGSON3R5)
+#define SNUMOPT         2
+#define DNUMOPT         2
+
+#define GEMM_DEFAULT_OFFSET_A 0
+#define GEMM_DEFAULT_OFFSET_B 0
+#define GEMM_DEFAULT_ALIGN 0x0ffffUL
+
+#define SGEMM_DEFAULT_UNROLL_N 8
+#define DGEMM_DEFAULT_UNROLL_N 8
+#define QGEMM_DEFAULT_UNROLL_N 2
+#define CGEMM_DEFAULT_UNROLL_N 4
+#define ZGEMM_DEFAULT_UNROLL_N 4
+#define XGEMM_DEFAULT_UNROLL_N 1
+
+#define SGEMM_DEFAULT_UNROLL_M 2
+#define DGEMM_DEFAULT_UNROLL_M 2
+#define QGEMM_DEFAULT_UNROLL_M 2
+#define CGEMM_DEFAULT_UNROLL_M 1
+#define ZGEMM_DEFAULT_UNROLL_M 1
+#define XGEMM_DEFAULT_UNROLL_M 1
+
+#define SGEMM_DEFAULT_P sgemm_p
+#define DGEMM_DEFAULT_P dgemm_p
+#define QGEMM_DEFAULT_P qgemm_p
+#define CGEMM_DEFAULT_P cgemm_p
+#define ZGEMM_DEFAULT_P zgemm_p
+#define XGEMM_DEFAULT_P xgemm_p
+
+#define SGEMM_DEFAULT_R sgemm_r
+#define DGEMM_DEFAULT_R dgemm_r
+#define QGEMM_DEFAULT_R qgemm_r
+#define CGEMM_DEFAULT_R cgemm_r
+#define ZGEMM_DEFAULT_R zgemm_r
+#define XGEMM_DEFAULT_R xgemm_r
+
+#define SGEMM_DEFAULT_Q 128
+#define DGEMM_DEFAULT_Q 128
+#define QGEMM_DEFAULT_Q 128
+#define CGEMM_DEFAULT_Q 128
+#define ZGEMM_DEFAULT_Q 128
+#define XGEMM_DEFAULT_Q 128
+
+#define SYMV_P  16
+#endif
+
 #if defined(P5600) || defined(MIPS1004K) || defined(MIPS24K) || defined(I6400) || defined(P6600) || defined(I6500)
 #define SNUMOPT  2
 #define DNUMOPT  2

test/CMakeLists.txt

@@ -2,6 +2,10 @@ include_directories(${PROJECT_SOURCE_DIR})
 include_directories(${PROJECT_BINARY_DIR})
 
 enable_language(Fortran)
+if (CMAKE_Fortran_COMPILER_ID STREQUAL GNU)
+        set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -fno-tree-vectorize")
+endif()
+
 
 if (BUILD_SINGLE)
 	list( APPEND OpenBLAS_Tests sblat1 sblat2 sblat3)

test/Makefile

@@ -1,6 +1,8 @@
 TOPDIR	= ..
 include ../Makefile.system
-
+ifeq ($(F_COMPILER),GFORTRAN)
+        override FFLAGS += -fno-tree-vectorize
+endif
 
 ifeq ($(NOFORTRAN),1)
 all ::