Merge branch 'develop' of https://github.com/quickwritereader/OpenBLAS into develop

2019-04-29 08:57:44 +00:00 · 2019-04-29 08:57:44 +00:00 · 628b335e83
parent 0f105dd8a5 7c51cc8527
commit 628b335e83
197 changed files with 17904 additions and 7444 deletions
--- a/.travis.yml
+++ b/.travis.yml
@ -149,7 +149,7 @@ matrix:

    - &test-macos
      os: osx
-      osx_image: xcode8.3
+      osx_image: xcode10.1
      before_script:
        - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"
        - brew update
@ -160,6 +160,7 @@ matrix:
        - BTYPE="BINARY=64 INTERFACE64=1"

    - <<: *test-macos
+      osx_image: xcode8.3
      env:
        - BTYPE="BINARY=32"

--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@ -42,6 +42,19 @@ endif()

 #######

+if(MSVC AND MSVC_STATIC_CRT)
+    set(CompilerFlags
+            CMAKE_CXX_FLAGS
+            CMAKE_CXX_FLAGS_DEBUG
+            CMAKE_CXX_FLAGS_RELEASE
+            CMAKE_C_FLAGS
+            CMAKE_C_FLAGS_DEBUG
+            CMAKE_C_FLAGS_RELEASE
+            )
+    foreach(CompilerFlag ${CompilerFlags})
+      string(REPLACE "/MD" "/MT" ${CompilerFlag} "${${CompilerFlag}}")
+    endforeach()
+endif()

 message(WARNING "CMake support is experimental. It does not yet support all build options and may not produce the same Makefiles that OpenBLAS ships with.")

@ -62,10 +75,10 @@ endif ()

 set(SUBDIRS	${BLASDIRS})
 if (NOT NO_LAPACK)
-  list(APPEND SUBDIRS lapack)
  if(BUILD_RELAPACK)
    list(APPEND SUBDIRS relapack/src)
  endif()
+  list(APPEND SUBDIRS lapack)
 endif ()

 # set which float types we want to build for
@ -134,7 +147,7 @@ endif ()

 # Only generate .def for dll on MSVC and always produce pdb files for debug and release
 if(MSVC)
-  if (${CMAKE_MAJOR_VERSION}.${CMAKE_MINOR_VERSION} LESS 3.4)
+  if (${CMAKE_MAJOR_VERSION}.${CMAKE_MINOR_VERSION} VERSION_LESS 3.4)
    set(OpenBLAS_DEF_FILE "${PROJECT_BINARY_DIR}/openblas.def")
  endif()
  set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} /Zi")
@ -149,15 +162,9 @@ if (${DYNAMIC_ARCH})
  endforeach()
 endif ()

-# Only build shared libs for MSVC
-if (MSVC)
-  set(BUILD_SHARED_LIBS ON)
-endif()
-
-
 # add objects to the openblas lib
 add_library(${OpenBLAS_LIBNAME} ${LA_SOURCES} ${LAPACKE_SOURCES} ${RELA_SOURCES} ${TARGET_OBJS} ${OpenBLAS_DEF_FILE})
-target_include_directories(${OpenBLAS_LIBNAME} INTERFACE $<INSTALL_INTERFACE:include>)
+target_include_directories(${OpenBLAS_LIBNAME} INTERFACE $<INSTALL_INTERFACE:include/openblas${SUFFIX64}>)

 # Android needs to explicitly link against libm
 if(ANDROID)
@ -166,7 +173,7 @@ endif()

 # Handle MSVC exports
 if(MSVC AND BUILD_SHARED_LIBS)
-  if (${CMAKE_MAJOR_VERSION}.${CMAKE_MINOR_VERSION} LESS 3.4)
+  if (${CMAKE_MAJOR_VERSION}.${CMAKE_MINOR_VERSION} VERSION_LESS 3.4)
    include("${PROJECT_SOURCE_DIR}/cmake/export.cmake")
  else()
    # Creates verbose .def file (51KB vs 18KB)
@ -217,6 +224,14 @@ set_target_properties(${OpenBLAS_LIBNAME} PROPERTIES
  SOVERSION ${OpenBLAS_MAJOR_VERSION}
 )

+if (BUILD_SHARED_LIBS AND BUILD_RELAPACK)
+  if (NOT MSVC)
+    target_link_libraries(${OpenBLAS_LIBNAME} "-Wl,-allow-multiple-definition")
+  else()
+    target_link_libraries(${OpenBLAS_LIBNAME} "/FORCE:MULTIPLE")
+  endif()
+endif()
+
 if (BUILD_SHARED_LIBS AND NOT ${SYMBOLPREFIX}${SYMBOLSUFIX} STREQUAL "")
 if (NOT DEFINED ARCH)
  set(ARCH_IN "x86_64")
@ -314,7 +329,7 @@ install (FILES ${OPENBLAS_CONFIG_H} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
 if(NOT NOFORTRAN)
  message(STATUS "Generating f77blas.h in ${CMAKE_INSTALL_INCLUDEDIR}")

-  set(F77BLAS_H ${CMAKE_BINARY_DIR}/f77blas.h)
+  set(F77BLAS_H ${CMAKE_BINARY_DIR}/generated/f77blas.h)
  file(WRITE  ${F77BLAS_H} "#ifndef OPENBLAS_F77BLAS_H\n")
  file(APPEND ${F77BLAS_H} "#define OPENBLAS_F77BLAS_H\n")
  file(APPEND ${F77BLAS_H} "#include \"openblas_config.h\"\n")
@ -327,10 +342,11 @@ endif()
 if(NOT NO_CBLAS)
 	message (STATUS "Generating cblas.h in ${CMAKE_INSTALL_INCLUDEDIR}")

+	set(CBLAS_H ${CMAKE_BINARY_DIR}/generated/cblas.h)
 	file(READ ${CMAKE_CURRENT_SOURCE_DIR}/cblas.h CBLAS_H_CONTENTS)
 	string(REPLACE "common" "openblas_config" CBLAS_H_CONTENTS_NEW "${CBLAS_H_CONTENTS}")
-	file(WRITE ${CMAKE_BINARY_DIR}/cblas.tmp "${CBLAS_H_CONTENTS_NEW}")
-	install (FILES ${CMAKE_BINARY_DIR}/cblas.tmp DESTINATION ${CMAKE_INSTALL_INCLUDEDIR} RENAME cblas.h)
+	file(WRITE ${CBLAS_H} "${CBLAS_H_CONTENTS_NEW}")
+	install (FILES ${CBLAS_H} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
 endif()

 if(NOT NO_LAPACKE)
--- a/2
+++ b/2
@ -96,7 +96,7 @@ endif
 	@echo

 shared :
-ifndef NO_SHARED
+ifneq ($(NO_SHARED), 1)
 ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android Haiku))
 	@$(MAKE) -C exports so
 	@ln -fs $(LIBSONAME) $(LIBPREFIX).so
--- a/Makefile.arm64
+++ b/Makefile.arm64
@ -38,3 +38,8 @@ ifeq ($(CORE), THUNDERX2T99)
 CCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
 FCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
 endif
+
+ifeq ($(CORE), TSV110)
+CCOMMON_OPT += -march=armv8.2-a -mtune=tsv110
+FCOMMON_OPT += -march=armv8.2-a -mtune=tsv110
+endif
--- a/Makefile.install
+++ b/Makefile.install
@ -58,14 +58,14 @@ ifndef NO_LAPACKE
 endif

 #for install static library
-ifndef NO_STATIC
+ifneq ($(NO_STATIC),1)
 	@echo Copying the static library to $(DESTDIR)$(OPENBLAS_LIBRARY_DIR)
 	@install -pm644 $(LIBNAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
 	ln -fs $(LIBNAME) $(LIBPREFIX).$(LIBSUFFIX)
 endif
 #for install shared library
-ifndef NO_SHARED
+ifneq ($(NO_SHARED),1)
 	@echo Copying the shared library to $(DESTDIR)$(OPENBLAS_LIBRARY_DIR)
 ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android Haiku))
 	@install -pm755 $(LIBSONAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
@ -106,14 +106,14 @@ ifndef NO_LAPACKE
 endif

 #for install static library
-ifndef NO_STATIC
+ifneq ($(NO_STATIC),1)
 	@echo Copying the static library to $(DESTDIR)$(OPENBLAS_LIBRARY_DIR)
 	@installbsd -c -m 644 $(LIBNAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
 	ln -fs $(LIBNAME) $(LIBPREFIX).$(LIBSUFFIX)
 endif
 #for install shared library
-ifndef NO_SHARED
+ifneq ($(NO_SHARED),1)
 	@echo Copying the shared library to $(DESTDIR)$(OPENBLAS_LIBRARY_DIR)
 	@installbsd -c -m 755 $(LIBSONAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
@ -138,7 +138,7 @@ endif
 	@echo "SET(OpenBLAS_VERSION \"${VERSION}\")" > "$(DESTDIR)$(OPENBLAS_CMAKE_DIR)/$(OPENBLAS_CMAKE_CONFIG)"
 	@echo "SET(OpenBLAS_INCLUDE_DIRS ${OPENBLAS_INCLUDE_DIR})" >> "$(DESTDIR)$(OPENBLAS_CMAKE_DIR)/$(OPENBLAS_CMAKE_CONFIG)"

-ifndef NO_SHARED
+ifneq ($(NO_SHARED),1)
 #ifeq logical or
 ifeq ($(OSNAME), $(filter $(OSNAME),Linux FreeBSD NetBSD OpenBSD DragonFly))
 	@echo "SET(OpenBLAS_LIBRARIES ${OPENBLAS_LIBRARY_DIR}/$(LIBPREFIX).so)" >> "$(DESTDIR)$(OPENBLAS_CMAKE_DIR)/$(OPENBLAS_CMAKE_CONFIG)"
--- a/Makefile.rule
+++ b/Makefile.rule
@ -48,6 +48,8 @@ VERSION = 0.3.6.dev
 # HOSTCC = gcc

 # If you need 32bit binary, define BINARY=32, otherwise define BINARY=64
+# Please note that AVX is not available on 32-bit.
+# Setting BINARY=32 disables AVX/AVX2/AVX-512.
 # BINARY=64

 # About threaded BLAS. It will be automatically detected if you don't
@ -57,7 +59,7 @@ VERSION = 0.3.6.dev
 # USE_THREAD = 0

 # If you're going to use this library with OpenMP, please comment it in.
-# This flag is always set for POWER8. Don't modify the flag 
+# This flag is always set for POWER8. Don't set USE_OPENMP = 0 if you're targeting POWER8.
 # USE_OPENMP = 1

 # The OpenMP scheduler to use - by default this is "static" and you
@ -68,36 +70,45 @@ VERSION = 0.3.6.dev
 # allow you to select the scheduler from the environment variable OMP_SCHEDULE
 # CCOMMON_OPT += -DOMP_SCHED=dynamic

-# You can define maximum number of threads. Basically it should be
-# less than actual number of cores. If you don't specify one, it's
-# automatically detected by the the script.
+# You can define the maximum number of threads. Basically it should be less
+# than or equal to the number of CPU threads. If you don't specify one, it's
+# automatically detected by the build system.
+# If SMT (aka. HT) is enabled on the system, it may or may not be beneficial to 
+# restrict NUM_THREADS to the number of physical cores. By default, the automatic 
+# detection includes logical CPUs, thus allowing the use of SMT.
+# Users may opt at runtime to use less than NUM_THREADS threads.
+#
+# Note for package maintainers: you can build OpenBLAS with a large NUM_THREADS
+# value (eg. 32-256) if you expect your users to use that many threads. Due to the way
+# some internal structures are allocated, using a large NUM_THREADS value has a RAM
+# footprint penalty, even if users reduce the actual number of threads at runtime.
 # NUM_THREADS = 24

 # If you have enabled USE_OPENMP and your application would call
-# OpenBLAS's calculation API from multi threads, please comment it in.
-# This flag defines how many instances of OpenBLAS's calculation API can 
-# actually run in parallel. If more threads call OpenBLAS's calculation API,
+# OpenBLAS's calculation API from multiple threads, please comment this in.
+# This flag defines how many instances of OpenBLAS's calculation API can actually
+# run in parallel. If more than NUM_PARALLEL threads call OpenBLAS's calculation API,
 # they need to wait for the preceding API calls to finish or risk data corruption.
 # NUM_PARALLEL = 2

-# if you don't need to install the static library, please comment it in.
+# If you don't need to install the static library, please comment this in.
 # NO_STATIC = 1

-# if you don't need generate the shared library, please comment it in.
+# If you don't need to generate the shared library, please comment this in.
 # NO_SHARED = 1

-# If you don't need CBLAS interface, please comment it in.
+# If you don't need the CBLAS interface, please comment this in.
 # NO_CBLAS = 1

-# If you only want CBLAS interface without installing Fortran compiler,
-# please comment it in.
+# If you only want the CBLAS interface without installing a Fortran compiler,
+# please comment this in.
 # ONLY_CBLAS = 1

-# If you don't need LAPACK, please comment it in.
-# If you set NO_LAPACK=1, the library automatically sets NO_LAPACKE=1.
+# If you don't need LAPACK, please comment this in.
+# If you set NO_LAPACK=1, the build system automatically sets NO_LAPACKE=1.
 # NO_LAPACK = 1

-# If you don't need LAPACKE (C Interface to LAPACK), please comment it in.
+# If you don't need LAPACKE (C Interface to LAPACK), please comment this in.
 # NO_LAPACKE = 1

 # Build LAPACK Deprecated functions since LAPACK 3.6.0
@ -106,7 +117,7 @@ BUILD_LAPACK_DEPRECATED = 1
 # Build RecursiveLAPACK on top of LAPACK
 # BUILD_RELAPACK = 1

-# If you want to use legacy threaded Level 3 implementation.
+# If you want to use the legacy threaded Level 3 implementation.
 # USE_SIMPLE_THREADED_LEVEL3 = 1

 # If you want to use the new, still somewhat experimental code that uses
@ -116,8 +127,8 @@ BUILD_LAPACK_DEPRECATED = 1
 # USE_TLS = 1

 # If you want to drive whole 64bit region by BLAS. Not all Fortran
-# compiler supports this. It's safe to keep comment it out if you
-# are not sure(equivalent to "-i8" option).
+# compilers support this. It's safe to keep this commented out if you
+# are not sure. (This is equivalent to the "-i8" ifort option).
 # INTERFACE64 = 1

 # Unfortunately most of kernel won't give us high quality buffer.
@ -125,10 +136,18 @@ BUILD_LAPACK_DEPRECATED = 1
 # but it will consume time. If you don't like it, you can disable one.
 NO_WARMUP = 1

-# If you want to disable CPU/Memory affinity on Linux.
+# Comment this in if you want to disable OpenBLAS's CPU/Memory affinity handling.
+# This feature is only implemented on Linux, and is always disabled on other platforms.
+# Enabling affinity handling may improve performance, especially on NUMA systems, but 
+# it may conflict with certain applications that also try to manage affinity.
+# This conflict can result in threads of the application calling OpenBLAS ending up locked
+# to the same core(s) as OpenBLAS, possibly binding all threads to a single core.
+# For this reason, affinity handling is disabled by default. Can be safely enabled if nothing
+# else modifies affinity settings.
+# Note: enabling affinity has been known to cause problems with NumPy and R
 NO_AFFINITY = 1

-# if you are compiling for Linux and you have more than 16 numa nodes or more than 256 cpus
+# If you are compiling for Linux and you have more than 16 numa nodes or more than 256 cpus
 # BIGNUMA = 1

 # Don't use AVX kernel on Sandy Bridge. It is compatible with old compilers
@ -180,7 +199,7 @@ NO_AFFINITY = 1
 # been reported to be optimal for certain workloads (50 is the recommended value for Julia).
 # GEMM_MULTITHREAD_THRESHOLD = 4

-# If you need santy check by comparing reference BLAS. It'll be very
+# If you need sanity check by comparing results to reference BLAS. It'll be very
 # slow (Not implemented yet).
 # SANITY_CHECK = 1

--- a/Makefile.system
+++ b/Makefile.system
@ -95,6 +95,9 @@ endif
 ifeq ($(TARGET), ZEN)
 GETARCH_FLAGS := -DFORCE_BARCELONA
 endif
+ifeq ($(TARGET), ARMV8)
+GETARCH_FLAGS := -DFORCE_ARMV7
+endif
 endif


@ -152,7 +155,8 @@ GETARCH_FLAGS	+= -DNO_AVX
 endif

 ifeq ($(BINARY), 32)
-GETARCH_FLAGS	+= -DNO_AVX
+GETARCH_FLAGS	+= -DNO_AVX -DNO_AVX2 -DNO_AVX512
+NO_AVX512 = 1
 endif

 ifeq ($(NO_AVX2), 1)
--- a/Makefile.zarch
+++ b/Makefile.zarch
@ -4,3 +4,7 @@ CCOMMON_OPT += -march=z13 -mzvector
 FCOMMON_OPT += -march=z13 -mzvector
 endif

+ifeq ($(CORE), Z14)
+CCOMMON_OPT += -march=z14 -mzvector
+FCOMMON_OPT += -march=z14 -mzvector
+endif
--- a/TargetList.txt
+++ b/TargetList.txt
@ -91,7 +91,9 @@ CORTEXA73
 FALKOR
 THUNDERX
 THUNDERX2T99
+TSV110

 9.System Z:
 ZARCH_GENERIC
 Z13
+Z14
--- a/appveyor.yml
+++ b/appveyor.yml
@ -53,9 +53,9 @@ before_build:
  - ps: if (-Not (Test-Path .\build)) { mkdir build }
  - cd build
  - if [%COMPILER%]==[cl] cmake -G "Visual Studio 15 2017 Win64" ..
-  - if [%WITH_FORTRAN%]==[no] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl ..
+  - if [%WITH_FORTRAN%]==[no] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl -DMSVC_STATIC_CRT=ON ..
  - if [%WITH_FORTRAN%]==[yes] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl -DCMAKE_Fortran_COMPILER=flang -DBUILD_WITHOUT_LAPACK=no -DNOFORTRAN=0 ..
-  - if [%DYNAMIC_ARCH%]==[ON] cmake -DDYNAMIC_ARCH=ON ..
+  - if [%DYNAMIC_ARCH%]==[ON] cmake -DDYNAMIC_ARCH=ON -DDYNAMIC_LIST='CORE2;NEHALEM;SANDYBRIDGE;BULLDOZER;HASWELL' ..

 build_script:
  - cmake --build .
--- a/benchmark/scripts/R/deig.R
+++ b/benchmark/scripts/R/deig.R
@ -2,6 +2,8 @@

 argv <- commandArgs(trailingOnly = TRUE)

+if (!is.null(options("matprod")[[1]])) options(matprod = "blas")
+
 nfrom <- 128
 nto <- 2048
 nstep <- 128
@ -19,7 +21,6 @@ if (length(argv) > 0) {
      loops <- as.numeric(argv[z])
    }
  }
-
 }

 p <- Sys.getenv("OPENBLAS_LOOPS")
@ -27,29 +28,21 @@ if (p != "") {
  loops <- as.numeric(p)
 }

-
-cat(sprintf(
-  "From %.0f To %.0f Step=%.0f Loops=%.0f\n",
-  nfrom,
-  nto,
-  nstep,
-  loops
-))
+cat(sprintf("From %.0f To %.0f Step=%.0f Loops=%.0f\n", nfrom, nto, nstep, loops))
 cat(sprintf("      SIZE             Flops                   Time\n"))

 n <- nfrom
 while (n <= nto) {
-  A <- matrix(rnorm(n * n), ncol = n, nrow = n)
+  A <- matrix(rnorm(n * n), nrow = n)
  ev <- 0
  z <- system.time(for (l in 1:loops) {
    ev <- eigen(A)
  })

-  mflops <- (26.66 * n * n * n) * loops / (z[3] * 1.0e6)
+  mflops <- (26.66 * n * n * n) * loops / (z[3] * 1e+06)

  st <- sprintf("%.0fx%.0f :", n, n)
  cat(sprintf("%20s %10.2f MFlops %10.6f sec\n", st, mflops, z[3]))

  n <- n + nstep
-
 }
--- a/benchmark/scripts/R/dgemm.R
+++ b/benchmark/scripts/R/dgemm.R
@ -2,6 +2,8 @@

 argv <- commandArgs(trailingOnly = TRUE)

+if (!is.null(options("matprod")[[1]])) options(matprod = "blas")
+
 nfrom <- 128
 nto <- 2048
 nstep <- 128
@ -19,7 +21,6 @@ if (length(argv) > 0) {
      loops <- as.numeric(argv[z])
    }
  }
-
 }

 p <- Sys.getenv("OPENBLAS_LOOPS")
@ -27,26 +28,13 @@ if (p != "") {
  loops <- as.numeric(p)
 }

-
-cat(sprintf(
-  "From %.0f To %.0f Step=%.0f Loops=%.0f\n",
-  nfrom,
-  nto,
-  nstep,
-  loops
-))
+cat(sprintf("From %.0f To %.0f Step=%.0f Loops=%.0f\n", nfrom, nto, nstep, loops))
 cat(sprintf("      SIZE             Flops                   Time\n"))

 n <- nfrom
 while (n <= nto) {
-  A <- matrix(runif(n * n),
-              ncol = n,
-              nrow = n,
-              byrow = TRUE)
-  B <- matrix(runif(n * n),
-              ncol = n,
-              nrow = n,
-              byrow = TRUE)
+  A <- matrix(runif(n * n), nrow = n)
+  B <- matrix(runif(n * n), nrow = n)
  C <- 1

  z <- system.time(for (l in 1:loops) {
@ -54,11 +42,10 @@ while (n <= nto) {
    l <- l + 1
  })

-  mflops <- (2.0 * n * n * n) * loops / (z[3] * 1.0e6)
+  mflops <- (2.0 * n * n * n) * loops / (z[3] * 1e+06)

  st <- sprintf("%.0fx%.0f :", n, n)
  cat(sprintf("%20s %10.2f MFlops %10.6f sec\n", st, mflops, z[3]))

  n <- n + nstep
-
 }
--- a/benchmark/scripts/R/dsolve.R
+++ b/benchmark/scripts/R/dsolve.R
@ -2,6 +2,8 @@

 argv <- commandArgs(trailingOnly = TRUE)

+if (!is.null(options("matprod")[[1]])) options(matprod = "blas")
+
 nfrom <- 128
 nto <- 2048
 nstep <- 128
@ -19,7 +21,6 @@ if (length(argv) > 0) {
      loops <- as.numeric(argv[z])
    }
  }
-
 }

 p <- Sys.getenv("OPENBLAS_LOOPS")
@ -27,31 +28,22 @@ if (p != "") {
  loops <- as.numeric(p)
 }

-
-cat(sprintf(
-  "From %.0f To %.0f Step=%.0f Loops=%.0f\n",
-  nfrom,
-  nto,
-  nstep,
-  loops
-))
+cat(sprintf("From %.0f To %.0f Step=%.0f Loops=%.0f\n", nfrom, nto, nstep, loops))
 cat(sprintf("      SIZE             Flops                   Time\n"))

 n <- nfrom
 while (n <= nto) {
-  A <- matrix(rnorm(n * n), ncol = n, nrow = n)
-  B <- matrix(rnorm(n * n), ncol = n, nrow = n)
+  A <- matrix(rnorm(n * n), nrow = n)
+  B <- matrix(rnorm(n * n), nrow = n)

  z <- system.time(for (l in 1:loops) {
    solve(A, B)
  })

-  mflops <-
-    (2.0 / 3.0 * n * n * n + 2.0 * n * n * n) * loops / (z[3] * 1.0e6)
+  mflops <- (8.0 / 3 * n * n * n) * loops / (z[3] * 1e+06)

  st <- sprintf("%.0fx%.0f :", n, n)
  cat(sprintf("%20s %10.2f MFlops %10.6f sec\n", st, mflops, z[3]))

  n <- n + nstep
-
 }
--- a/85
+++ b/85
@ -1,7 +1,7 @@
 #!/usr/bin/perl

-use File::Basename;
-use File::Temp qw(tempfile);
+#use File::Basename;
+# use File::Temp qw(tempfile);

 # Checking cross compile
 $hostos   = `uname -s | sed -e s/\-.*//`;    chop($hostos);
@ -12,7 +12,7 @@ $hostarch = "arm64" if ($hostarch eq "aarch64");
 $hostarch = "power" if ($hostarch =~ /^(powerpc|ppc).*/);
 $hostarch = "zarch" if ($hostarch eq "s390x");

-$tmpf = new File::Temp( UNLINK => 1 );
+#$tmpf = new File::Temp( UNLINK => 1 );
 $binary = $ENV{"BINARY"};

 $makefile = shift(@ARGV);
@ -31,12 +31,25 @@ if ($?) {

 $cross_suffix = "";

-if (dirname($compiler_name) ne ".") {
-    $cross_suffix .= dirname($compiler_name) . "/";
-}
+eval "use File::Basename";
+if ($@){ 
+    warn "could not load PERL module File::Basename, emulating its functionality";
+    my $dirnam = substr($compiler_name, 0, rindex($compiler_name, "/")-1 );
+    if ($dirnam ne ".") {
+	$cross_suffix .= $dirnam . "/";
+    }
+    my $basnam = substr($compiler_name, rindex($compiler_name,"/")+1, length($compiler_name)-rindex($compiler_name,"/")-1);
+	if ($basnam =~ /([^\s]*-)(.*)/) {
+	$cross_suffix .= $1;
+    }
+} else {
+    if (dirname($compiler_name) ne ".") {
+	$cross_suffix .= dirname($compiler_name) . "/";
+    }

-if (basename($compiler_name) =~ /([^\s]*-)(.*)/) {
-    $cross_suffix .= $1;
+    if (basename($compiler_name) =~ /([^\s]*-)(.*)/) {
+	$cross_suffix .= $1;
+    }
 }

 $compiler = "";
@ -171,20 +184,26 @@ if ($?) {

 $have_msa = 0;
 if (($architecture eq "mips") || ($architecture eq "mips64")) {
-    $code = '"addvi.b $w0, $w1, 1"';
-    $msa_flags = "-mmsa -mfp64 -msched-weight -mload-store-pairs";
-    print $tmpf "#include <msa.h>\n\n";
-    print $tmpf "void main(void){ __asm__ volatile($code); }\n";
-
-    $args = "$msa_flags -o $tmpf.o -x c $tmpf";
-    my @cmd = ("$compiler_name $args");
-    system(@cmd) == 0;
-    if ($? != 0) {
-        $have_msa = 0;
+    eval "use File::Temp qw(tempfile)";
+    if ($@){ 
+	warn "could not load PERL module File::Temp, so could not check MSA capatibility";
    } else {
-        $have_msa = 1;
+	$tmpf = new File::Temp( UNLINK => 1 );
+	$code = '"addvi.b $w0, $w1, 1"';
+	$msa_flags = "-mmsa -mfp64 -msched-weight -mload-store-pairs";
+	print $tmpf "#include <msa.h>\n\n";
+	print $tmpf "void main(void){ __asm__ volatile($code); }\n";
+
+	$args = "$msa_flags -o $tmpf.o -x c $tmpf";
+	my @cmd = ("$compiler_name $args");
+	system(@cmd) == 0;
+	if ($? != 0) {
+	    $have_msa = 0;
+	} else {
+	    $have_msa = 1;
+	}
+	unlink("$tmpf.o");
    }
-    unlink("$tmpf.o");
 }

 $architecture = x86    if ($data =~ /ARCH_X86/);
@ -204,17 +223,25 @@ $binformat    = bin64  if ($data =~ /BINARY_64/);

 $no_avx512= 0;
 if (($architecture eq "x86") || ($architecture eq "x86_64")) {
-    $code = '"vbroadcastss -4 * 4(%rsi), %zmm2"';
-    print $tmpf "#include <immintrin.h>\n\nint main(void){ __asm__ volatile($code); }\n";
-    $args = " -march=skylake-avx512 -o $tmpf.o -x c $tmpf";
-    my @cmd = ("$compiler_name $args >/dev/null 2>/dev/null");
-    system(@cmd) == 0;
-    if ($? != 0) {
-	$no_avx512 = 1;
-    } else {
+    eval "use File::Temp qw(tempfile)";
+    if ($@){ 
+	warn "could not load PERL module File::Temp, so could not check compiler compatibility with AVX512";
 	$no_avx512 = 0;
+    } else {
+#	$tmpf = new File::Temp( UNLINK => 1 );
+	($fh,$tmpf) = tempfile( UNLINK => 1 );
+	$code = '"vbroadcastss -4 * 4(%rsi), %zmm2"';
+	print $tmpf "#include <immintrin.h>\n\nint main(void){ __asm__ volatile($code); }\n";
+	$args = " -march=skylake-avx512 -c -o $tmpf.o -x c $tmpf";
+	my @cmd = ("$compiler_name $args >/dev/null 2>/dev/null");
+	system(@cmd) == 0;
+	if ($? != 0) {
+	    $no_avx512 = 1;
+	} else {
+	    $no_avx512 = 0;
+	}
+	unlink("tmpf.o");
    }
-    unlink("tmpf.o");
 }

 $data = `$compiler_name -S ctest1.c && grep globl ctest1.s | head -n 1 && rm -f ctest1.s`;
--- a/cmake/arch.cmake
+++ b/cmake/arch.cmake
@ -74,6 +74,9 @@ if (DYNAMIC_ARCH)
    if (NOT NO_AVX512)
      set(DYNAMIC_CORE ${DYNAMIC_CORE} SKYLAKEX)
    endif ()
+    if (DYNAMIC_LIST)
+	set(DYNAMIC_CORE PRESCOTT ${DYNAMIC_LIST})
+    endif ()
  endif ()

  if (NOT DYNAMIC_CORE)
--- a/cmake/system.cmake
+++ b/cmake/system.cmake
@ -39,6 +39,9 @@ if (DEFINED BINARY AND DEFINED TARGET AND BINARY EQUAL 32)
  if (${TARGET} STREQUAL "BULLDOZER" OR ${TARGET} STREQUAL "PILEDRIVER" OR ${TARGET} STREQUAL "ZEN")
    set(TARGET "BARCELONA")
  endif ()
+  if (${TARGET} STREQUAL "ARMV8" OR ${TARGET} STREQUAL "CORTEXA57" OR ${TARGET} STREQUAL "CORTEXA53")
+    set(TARGET "ARMV7")
+  endif ()
 endif ()

 if (DEFINED TARGET)
@ -184,6 +187,13 @@ if (DYNAMIC_ARCH)
  endif ()
 endif ()

+if (DYNAMIC_LIST)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_LIST")
+  foreach(DCORE ${DYNAMIC_LIST})
+    set(CCOMMON_OPT "${CCOMMON_OPT} -DDYN_${DCORE}")
+  endforeach ()
+endif ()
+
 if (NO_LAPACK)
  set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_LAPACK")
  #Disable LAPACK C interface
--- a/cmake/system_check.cmake
+++ b/cmake/system_check.cmake
@ -39,7 +39,11 @@ elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "ppc.*|power.*|Power.*")
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "mips64.*")
  set(MIPS64 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "amd64.*|x86_64.*|AMD64.*")
-  set(X86_64 1)
+  if("${CMAKE_SIZEOF_VOID_P}" EQUAL "8")
+    set(X86_64 1)
+  else()
+    set(X86 1)
+  endif()
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "i686.*|i386.*|x86.*|amd64.*|AMD64.*")
  set(X86 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(arm.*|ARM.*)")
@ -78,7 +82,7 @@ endif()

 if (X86_64 OR X86)
  file(WRITE ${PROJECT_BINARY_DIR}/avx512.tmp "#include <immintrin.h>\n\nint main(void){ __asm__ volatile(\"vbroadcastss -4 * 4(%rsi), %zmm2\"); }")
-execute_process(COMMAND ${CMAKE_C_COMPILER} -march=skylake-avx512 -v -o ${PROJECT_BINARY_DIR}/avx512.o -x c ${PROJECT_BINARY_DIR}/avx512.tmp OUTPUT_QUIET ERROR_QUIET RESULT_VARIABLE NO_AVX512)
+execute_process(COMMAND ${CMAKE_C_COMPILER} -march=skylake-avx512 -c -v -o ${PROJECT_BINARY_DIR}/avx512.o -x c ${PROJECT_BINARY_DIR}/avx512.tmp OUTPUT_QUIET ERROR_QUIET RESULT_VARIABLE NO_AVX512)
 if (NO_AVX512 EQUAL 1)
 set (CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX512")
 endif()
--- a/common.h
+++ b/common.h
@ -444,7 +444,7 @@ please https://github.com/xianyi/OpenBLAS/issues/246
 typedef char env_var_t[MAX_PATH];
 #define readenv(p, n) 0
 #else
-#ifdef OS_WINDOWS
+#if defined(OS_WINDOWS) && !defined(OS_CYGWIN_NT)
 typedef char env_var_t[MAX_PATH];
 #define readenv(p, n) GetEnvironmentVariable((LPCTSTR)(n), (LPTSTR)(p), sizeof(p))
 #else
--- a/common_power.h
+++ b/common_power.h
@ -241,7 +241,7 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #define HAVE_PREFETCH
 #endif

-#if defined(POWER3) || defined(POWER6) || defined(PPCG4) || defined(CELL) || defined(POWER8)  || defined(POWER9)
+#if defined(POWER3) || defined(POWER6) || defined(PPCG4) || defined(CELL) || defined(POWER8) || defined(POWER9) || ( defined(PPC970) && defined(OS_DARWIN) )
 #define DCBT_ARG	0
 #else
 #define DCBT_ARG	8
@ -598,9 +598,14 @@ REALNAME:;\
 #ifndef __64BIT__
 #define PROLOGUE \
 	.machine "any";\
+	.toc;\
 	.globl .REALNAME;\
+	.globl REALNAME;\
+	.csect REALNAME[DS],3;\
+REALNAME:;\
+	.long .REALNAME, TOC[tc0], 0;\
 	.csect .text[PR],5;\
-.REALNAME:;
+.REALNAME:

 #define EPILOGUE \
 _section_.text:;\
@ -611,9 +616,14 @@ _section_.text:;\

 #define PROLOGUE \
 	.machine "any";\
+	.toc;\
 	.globl .REALNAME;\
+	.globl REALNAME;\
+	.csect REALNAME[DS],3;\
+REALNAME:;\
+	.llong .REALNAME, TOC[tc0], 0;\
 	.csect .text[PR], 5;\
-.REALNAME:;
+.REALNAME:

 #define EPILOGUE \
 _section_.text:;\
--- a/common_x86.h
+++ b/common_x86.h
@ -187,7 +187,7 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 	
  y = blas_quick_divide_table[y];

-  __asm__ __volatile__  ("mull %0" :"=d" (result) :"a"(x), "0" (y));
+  __asm__ __volatile__  ("mull %0" :"=d" (result), "+a"(x): "0" (y));

  return result;
 #endif
--- a/common_x86_64.h
+++ b/common_x86_64.h
@ -210,7 +210,7 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 	
  y = blas_quick_divide_table[y];

-  __asm__ __volatile__  ("mull %0" :"=d" (result) :"a"(x), "0" (y));
+  __asm__ __volatile__  ("mull %0" :"=d" (result), "+a"(x) : "0" (y));

  return result;
 }
--- a/cpuid_arm64.c
+++ b/cpuid_arm64.c
@ -39,6 +39,8 @@
 // Cavium
 #define CPU_THUNDERX      7
 #define CPU_THUNDERX2T99  8
+//Hisilicon
+#define CPU_TSV110        9

 static char *cpuname[] = {
  "UNKNOWN",
@ -49,7 +51,8 @@ static char *cpuname[] = {
  "CORTEXA73",
  "FALKOR",
  "THUNDERX",
-  "THUNDERX2T99"
+  "THUNDERX2T99",
+  "TSV110"
 };

 static char *cpuname_lower[] = {
@ -61,7 +64,8 @@ static char *cpuname_lower[] = {
  "cortexa73",
  "falkor",
  "thunderx",
-  "thunderx2t99"
+  "thunderx2t99",
+  "tsv110"
 };

 int get_feature(char *search)
@ -145,6 +149,9 @@ int detect(void)
 			return CPU_THUNDERX;
    else if (strstr(cpu_implementer, "0x43") && strstr(cpu_part, "0x0af"))
 			return CPU_THUNDERX2T99;
+    // HiSilicon
+    else if (strstr(cpu_implementer, "0x48") && strstr(cpu_part, "0xd01"))
+                        return CPU_TSV110;
 	}

 	p = (char *) NULL ;
@ -286,6 +293,21 @@ void get_cpuconfig(void)
 			printf("#define DTB_DEFAULT_ENTRIES  64       \n");
 			printf("#define DTB_SIZE             4096     \n");
 			break;
+			
+		case CPU_TSV110:
+			printf("#define TSV110                        \n");
+			printf("#define L1_CODE_SIZE         65536    \n");
+			printf("#define L1_CODE_LINESIZE     64       \n");
+			printf("#define L1_CODE_ASSOCIATIVE  4        \n");
+			printf("#define L1_DATA_SIZE         65536    \n");
+			printf("#define L1_DATA_LINESIZE     64       \n");
+			printf("#define L1_DATA_ASSOCIATIVE  4        \n");
+			printf("#define L2_SIZE              524228   \n");
+			printf("#define L2_LINESIZE          64       \n");
+			printf("#define L2_ASSOCIATIVE       8        \n");
+			printf("#define DTB_DEFAULT_ENTRIES  64       \n");
+			printf("#define DTB_SIZE             4096     \n");
+			break;	
 	}
 }

--- a/cpuid_x86.c
+++ b/cpuid_x86.c
@ -228,7 +228,7 @@ int support_avx2(){
 }

 int support_avx512(){
-#ifndef NO_AVX512
+#if !defined(NO_AVX) && !defined(NO_AVX512)
  int eax, ebx, ecx, edx;
  int ret=0;

@ -1359,6 +1359,8 @@ int get_cpuname(void){
 	    return CPUTYPE_NEHALEM;
 	case 12:
 	    // Apollo Lake
+	case 15:
+	    // Denverton		
 	    return CPUTYPE_NEHALEM;
 	}
 	break;
@ -1378,7 +1380,7 @@ int get_cpuname(void){
      case 9:
      case 8:      
        switch (model) {
-	case 14: // Kaby Lake
+	case 14: // Kaby Lake and refreshes
          if(support_avx2())
            return CPUTYPE_HASWELL;
          if(support_avx())
--- a/cpuid_zarch.c
+++ b/cpuid_zarch.c
@ -27,9 +27,9 @@

 #include <string.h>

-#define CPU_GENERIC    	0
-#define CPU_Z13       	1
-#define CPU_Z14       	2
+#define CPU_GENERIC     0
+#define CPU_Z13         1
+#define CPU_Z14         2

 static char *cpuname[] = {
  "ZARCH_GENERIC",
@ -64,10 +64,8 @@ int detect(void)

  if (strstr(p, "2964")) return CPU_Z13;
  if (strstr(p, "2965")) return CPU_Z13;
-
-  /* detect z14, but fall back to z13 */
-  if (strstr(p, "3906")) return CPU_Z13;
-  if (strstr(p, "3907")) return CPU_Z13;
+  if (strstr(p, "3906")) return CPU_Z14;
+  if (strstr(p, "3907")) return CPU_Z14;

  return CPU_GENERIC;
 }
@ -116,7 +114,14 @@ void get_cpuconfig(void)
 	  break;
 	case CPU_Z14:
 	  printf("#define Z14\n");
+	  printf("#define L1_DATA_SIZE 131072\n");
+	  printf("#define L1_DATA_LINESIZE 256\n");
+	  printf("#define L1_DATA_ASSOCIATIVE 8\n");
+	  printf("#define L2_SIZE 4194304\n");
+	  printf("#define L2_LINESIZE 256\n");
+	  printf("#define L2_ASSOCIATIVE 8\n");
 	  printf("#define DTB_DEFAULT_ENTRIES 64\n");
+	  printf("#define DTB_SIZE 4096\n");
 	  break;
 	}
 }
--- a/ctest.c
+++ b/ctest.c
@ -113,7 +113,7 @@ ARCH_X86
 ARCH_X86_64
 #endif

-#if defined(__powerpc___) || defined(__PPC__) || defined(_POWER)
+#if defined(__powerpc___) || defined(__PPC__) || defined(_POWER) || defined(__POWERPC__)
 ARCH_POWER
 #endif

--- a/driver/level2/trmv_thread.c
+++ b/driver/level2/trmv_thread.c
@ -346,7 +346,7 @@ int CNAME(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG incx, FLOAT *bu

    range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
    range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
-    if (range_n[num_cpu] > m) range_n[num_cpu] = m;
+    if (range_n[num_cpu] > m * num_cpu) range_n[num_cpu] = m * num_cpu;

    queue[num_cpu].mode    = mode;
    queue[num_cpu].routine = trmv_kernel;
@ -386,7 +386,7 @@ int CNAME(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG incx, FLOAT *bu

    range_m[num_cpu + 1] = range_m[num_cpu] + width;
    range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
-    if (range_n[num_cpu] > m) range_n[num_cpu] = m;
+    if (range_n[num_cpu] > m * num_cpu) range_n[num_cpu] = m * num_cpu;

    queue[num_cpu].mode    = mode;
    queue[num_cpu].routine = trmv_kernel;
--- a/driver/others/blas_server_win32.c
+++ b/driver/others/blas_server_win32.c
@ -461,13 +461,18 @@ int BLASFUNC(blas_thread_shutdown)(void){
    SetEvent(pool.killed);

    for(i = 0; i < blas_num_threads - 1; i++){
+      // Could also just use WaitForMultipleObjects
      WaitForSingleObject(blas_threads[i], 5);  //INFINITE);
 #ifndef OS_WINDOWSSTORE
 // TerminateThread is only available with WINAPI_DESKTOP and WINAPI_SYSTEM not WINAPI_APP in UWP
      TerminateThread(blas_threads[i],0);
 #endif
+      CloseHandle(blas_threads[i]);
    }

+    CloseHandle(pool.filled);
+    CloseHandle(pool.killed);
+
    blas_server_avail = 0;
  }

--- a/driver/others/dynamic.c
+++ b/driver/others/dynamic.c
@ -322,7 +322,7 @@ int support_avx2(){
 }

 int support_avx512(){
-#ifndef NO_AVX512
+#if !defined(NO_AVX) && !defined(NO_AVX512)
  int eax, ebx, ecx, edx;
  int ret=0;

@ -566,8 +566,8 @@ static gotoblas_t *get_coretype(void){
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
 	}
-	//Apollo Lake
-	if (model == 12) { 
+	//Apollo Lake or Denverton
+	if (model == 12 || model == 15) { 
 	  return &gotoblas_NEHALEM;
 	}	
 	return NULL;
--- a/driver/others/memory.c
+++ b/driver/others/memory.c
@ -198,45 +198,68 @@ int get_num_procs(void);
 #else
 int get_num_procs(void) {
  static int nums = 0;
-cpu_set_t *cpusetp;
-size_t size;
-int ret;
-int i,n;
+  cpu_set_t cpuset,*cpusetp;
+  size_t size;
+  int ret;
+
+#if defined(__GLIBC_PREREQ)
+#if !__GLIBC_PREREQ(2, 7)
+  int i;
+#if !__GLIBC_PREREQ(2, 6)
+  int n;
+#endif
+#endif
+#endif

  if (!nums) nums = sysconf(_SC_NPROCESSORS_CONF);
 #if !defined(OS_LINUX)
-     return nums;
+  return nums;
 #endif

 #if !defined(__GLIBC_PREREQ)
-   return nums;
+  return nums;
 #else
 #if !__GLIBC_PREREQ(2, 3)
-   return nums;
+  return nums;
 #endif

 #if !__GLIBC_PREREQ(2, 7)
-  ret = sched_getaffinity(0,sizeof(cpu_set_t), cpusetp);
+  ret = sched_getaffinity(0,sizeof(cpuset), &cpuset);
  if (ret!=0) return nums;
  n=0;
  #if !__GLIBC_PREREQ(2, 6)
  for (i=0;i<nums;i++)
-     if (CPU_ISSET(i,cpusetp)) n++;
+     if (CPU_ISSET(i,cpuset)) n++;
  nums=n;
  #else
-  nums = CPU_COUNT(sizeof(cpu_set_t),cpusetp);
+  nums = CPU_COUNT(sizeof(cpuset),&cpuset);
  #endif
  return nums;
 #else
-  cpusetp = CPU_ALLOC(nums);
-  if (cpusetp == NULL) return nums;
-  size = CPU_ALLOC_SIZE(nums);
-  ret = sched_getaffinity(0,size,cpusetp);
-  if (ret!=0) return nums;
-  ret = CPU_COUNT_S(size,cpusetp);
-  if (ret > 0 && ret < nums) nums = ret;
-  CPU_FREE(cpusetp);
-  return nums;
+  if (nums >= CPU_SETSIZE) {
+    cpusetp = CPU_ALLOC(nums);
+      if (cpusetp == NULL) {
+        return nums;
+      }
+    size = CPU_ALLOC_SIZE(nums);
+    ret = sched_getaffinity(0,size,cpusetp);
+    if (ret!=0) {
+      CPU_FREE(cpusetp);
+      return nums;
+    }
+    ret = CPU_COUNT_S(size,cpusetp);
+    if (ret > 0 && ret < nums) nums = ret;	
+    CPU_FREE(cpusetp);
+    return nums;
+  } else {
+    ret = sched_getaffinity(0,sizeof(cpuset),&cpuset);
+    if (ret!=0) {
+      return nums;
+    }
+    ret = CPU_COUNT(&cpuset);
+    if (ret > 0 && ret < nums) nums = ret;	
+    return nums;
+  }
 #endif
 #endif
 }
@ -1290,6 +1313,13 @@ void blas_memory_free_nolock(void * map_address) {
  free(map_address);
 }

+#ifdef SMP
+void blas_thread_memory_cleanup(void) {
+    blas_memory_cleanup((void*)get_memory_table());
+}
+#endif
+
+
 void blas_shutdown(void){
 #ifdef SMP
  BLASFUNC(blas_thread_shutdown)();
@ -1299,7 +1329,7 @@ void blas_shutdown(void){
  /* Only cleanupIf we were built for threading and TLS was initialized */
  if (local_storage_key)
 #endif
-    blas_memory_cleanup((void*)get_memory_table());
+    blas_thread_memory_cleanup();

 #ifdef SEEK_ADDRESS
  base_address      = 0UL;
@ -1529,7 +1559,7 @@ BOOL APIENTRY DllMain(HMODULE hModule, DWORD  ul_reason_for_call, LPVOID lpReser
      break;
    case DLL_THREAD_DETACH:
 #if defined(SMP)
-      blas_memory_cleanup((void*)get_memory_table());
+      blas_thread_memory_cleanup();
 #endif
      break;
    case DLL_PROCESS_DETACH:
@ -1603,9 +1633,11 @@ void gotoblas_dummy_for_PGI(void) {
 #endif

 #else
+/* USE_TLS / COMPILE_TLS not set */
+
 #include <errno.h>

-#ifdef OS_WINDOWS
+#if defined(OS_WINDOWS) && !defined(OS_CYGWIN_NT)
 #define ALLOC_WINDOWS
 #ifndef MEM_LARGE_PAGES
 #define MEM_LARGE_PAGES  0x20000000
@ -1619,7 +1651,7 @@ void gotoblas_dummy_for_PGI(void) {
 #include <stdio.h>
 #include <fcntl.h>

-#ifndef OS_WINDOWS
+#if !defined(OS_WINDOWS) || defined(OS_CYGWIN_NT)
 #include <sys/mman.h>
 #ifndef NO_SYSV_IPC
 #include <sys/shm.h>
@ -1639,7 +1671,7 @@ void gotoblas_dummy_for_PGI(void) {
 #include <sys/resource.h>
 #endif

-#if defined(OS_FREEBSD) || defined(OS_DARWIN)
+#if defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN)
 #include <sys/sysctl.h>
 #include <sys/resource.h>
 #endif
@ -1678,9 +1710,12 @@ void gotoblas_dummy_for_PGI(void) {
 #elif (defined(OS_DARWIN) || defined(OS_SUNOS)) && defined(C_GCC)
 #define CONSTRUCTOR	__attribute__ ((constructor))
 #define DESTRUCTOR	__attribute__ ((destructor))
-#else
+#elif __GNUC__ && INIT_PRIORITY && ((GCC_VERSION >= 40300) || (CLANG_VERSION >= 20900))
 #define CONSTRUCTOR	__attribute__ ((constructor(101)))
 #define DESTRUCTOR	__attribute__ ((destructor(101)))
+#else
+#define CONSTRUCTOR	__attribute__ ((constructor))
+#define DESTRUCTOR	__attribute__ ((destructor))
 #endif

 #ifdef DYNAMIC_ARCH
@ -1704,45 +1739,70 @@ void goto_set_num_threads(int num_threads) {};
 int get_num_procs(void);
 #else
 int get_num_procs(void) {
+
  static int nums = 0;
-cpu_set_t *cpusetp;
-size_t size;
-int ret;
-int i,n;
+  cpu_set_t cpuset,*cpusetp;
+  size_t size;
+  int ret;
+
+#if defined(__GLIBC_PREREQ)
+#if !__GLIBC_PREREQ(2, 7)
+  int i;
+#if !__GLIBC_PREREQ(2, 6)
+  int n;
+#endif
+#endif
+#endif

  if (!nums) nums = sysconf(_SC_NPROCESSORS_CONF);
 #if !defined(OS_LINUX)
-     return nums;
+  return nums;
 #endif

 #if !defined(__GLIBC_PREREQ)
-   return nums;
+  return nums;
 #else
 #if !__GLIBC_PREREQ(2, 3)
-   return nums;
+  return nums;
 #endif

 #if !__GLIBC_PREREQ(2, 7)
-  ret = sched_getaffinity(0,sizeof(cpu_set_t), cpusetp);
+  ret = sched_getaffinity(0,sizeof(cpuset), &cpuset);
  if (ret!=0) return nums;
  n=0;
  #if !__GLIBC_PREREQ(2, 6)
  for (i=0;i<nums;i++)
-     if (CPU_ISSET(i,cpusetp)) n++;
+     if (CPU_ISSET(i,cpuset)) n++;
  nums=n;
  #else
-  nums = CPU_COUNT(sizeof(cpu_set_t),cpusetp);
+  nums = CPU_COUNT(sizeof(cpuset),&cpuset);
  #endif
  return nums;
 #else
-  cpusetp = CPU_ALLOC(nums);
-  if (cpusetp == NULL) return nums;
-  size = CPU_ALLOC_SIZE(nums);
-  ret = sched_getaffinity(0,size,cpusetp);
-  if (ret!=0) return nums;
-  nums = CPU_COUNT_S(size,cpusetp);
-  CPU_FREE(cpusetp);
-  return nums;
+  if (nums >= CPU_SETSIZE) {
+    cpusetp = CPU_ALLOC(nums);
+      if (cpusetp == NULL) {
+        return nums;
+      }
+    size = CPU_ALLOC_SIZE(nums);
+    ret = sched_getaffinity(0,size,cpusetp);
+    if (ret!=0) {
+      CPU_FREE(cpusetp);
+      return nums;
+    }
+    ret = CPU_COUNT_S(size,cpusetp);
+    if (ret > 0 && ret < nums) nums = ret;	
+    CPU_FREE(cpusetp);
+    return nums;
+  } else {
+    ret = sched_getaffinity(0,sizeof(cpuset),&cpuset);
+    if (ret!=0) {
+      return nums;
+    }
+    ret = CPU_COUNT(&cpuset);
+    if (ret > 0 && ret < nums) nums = ret;	
+    return nums;
+  }
 #endif
 #endif
 }
@ -1793,7 +1853,7 @@ int get_num_procs(void) {

 #endif

-#if defined(OS_FREEBSD)
+#if defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_DRAGONFLY)

 int get_num_procs(void) {

@ -1870,7 +1930,7 @@ void openblas_fork_handler()
  //   http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60035
  // In the mean time build with USE_OPENMP=0 or link against another
  // implementation of OpenMP.
-#if !(defined(OS_WINDOWS) || defined(OS_ANDROID)) && defined(SMP_SERVER)
+#if !((defined(OS_WINDOWS) && !defined(OS_CYGWIN_NT)) || defined(OS_ANDROID)) && defined(SMP_SERVER)
  int err;
  err = pthread_atfork ((void (*)(void)) BLASFUNC(blas_thread_shutdown), NULL, NULL);
  if(err != 0)
@ -1883,7 +1943,7 @@ 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_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
  int max_num;
 #endif
  int blas_goto_num   = 0;
@ -1891,11 +1951,11 @@ 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_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
  max_num = get_num_procs();
 #endif

-  blas_goto_num = 0;
+  // blas_goto_num = 0;
 #ifndef USE_OPENMP
  blas_goto_num=openblas_num_threads_env();
  if (blas_goto_num < 0) blas_goto_num = 0;
@ -1907,7 +1967,7 @@ int blas_get_cpu_number(void){

 #endif

-  blas_omp_num = 0;
+  // blas_omp_num = 0;
  blas_omp_num=openblas_omp_num_threads_env();
  if (blas_omp_num < 0) blas_omp_num = 0;

@ -1915,7 +1975,7 @@ 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_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
  if (blas_num_threads > max_num) blas_num_threads = max_num;
 #endif

@ -2002,11 +2062,15 @@ static void *alloc_mmap(void *address){
  }

  if (map_address != (void *)-1) {
+#if defined(SMP) && !defined(USE_OPENMP)
    LOCK_COMMAND(&alloc_lock);
+#endif    
    release_info[release_pos].address = map_address;
    release_info[release_pos].func    = alloc_mmap_free;
    release_pos ++;
+#if defined(SMP) && !defined(USE_OPENMP)
    UNLOCK_COMMAND(&alloc_lock);
+#endif    
  }

 #ifdef OS_LINUX
@ -2148,14 +2212,18 @@ static void *alloc_mmap(void *address){
 #if defined(OS_LINUX) && !defined(NO_WARMUP)
  }
 #endif
-  LOCK_COMMAND(&alloc_lock);

  if (map_address != (void *)-1) {
+#if defined(SMP) && !defined(USE_OPENMP)
+    LOCK_COMMAND(&alloc_lock);
+#endif
    release_info[release_pos].address = map_address;
    release_info[release_pos].func    = alloc_mmap_free;
    release_pos ++;
+#if defined(SMP) && !defined(USE_OPENMP)
+    UNLOCK_COMMAND(&alloc_lock);
+#endif
  }
-  UNLOCK_COMMAND(&alloc_lock);

  return map_address;
 }
@ -2523,7 +2591,7 @@ void *blas_memory_alloc(int procpos){

  int position;
 #if defined(WHEREAMI) && !defined(USE_OPENMP)
-  int mypos;
+  int mypos = 0;
 #endif

  void *map_address;
@ -2554,6 +2622,11 @@ void *blas_memory_alloc(int procpos){
    NULL,
  };
  void *(**func)(void *address);
+
+#if defined(USE_OPENMP)
+  if (!memory_initialized) {
+#endif
+
  LOCK_COMMAND(&alloc_lock);

  if (!memory_initialized) {
@ -2589,6 +2662,9 @@ void *blas_memory_alloc(int procpos){

  }
  UNLOCK_COMMAND(&alloc_lock);
+#if defined(USE_OPENMP)
+  }
+#endif

 #ifdef DEBUG
  printf("Alloc Start ...\n");
@ -2603,13 +2679,17 @@ void *blas_memory_alloc(int procpos){

  do {
    if (!memory[position].used && (memory[position].pos == mypos)) {
+#if defined(SMP) && !defined(USE_OPENMP)
      LOCK_COMMAND(&alloc_lock);
-//      blas_lock(&memory[position].lock);
-
+#else      
+      blas_lock(&memory[position].lock);
+#endif
      if (!memory[position].used) goto allocation;
-
+#if defined(SMP) && !defined(USE_OPENMP)
      UNLOCK_COMMAND(&alloc_lock);
-//      blas_unlock(&memory[position].lock);
+#else
+      blas_unlock(&memory[position].lock);
+#endif      
    }

    position ++;
@ -2621,21 +2701,26 @@ void *blas_memory_alloc(int procpos){

  position = 0;

+#if defined(SMP) && !defined(USE_OPENMP)
  LOCK_COMMAND(&alloc_lock);
+#endif
  do {
-/*    if (!memory[position].used) { */
-/*      blas_lock(&memory[position].lock);*/
-
+#if defined(USE_OPENMP)	  
+    if (!memory[position].used) { 
+      blas_lock(&memory[position].lock);
+#endif
      if (!memory[position].used) goto allocation;
      
-/*      blas_unlock(&memory[position].lock);*/
-/*    } */
-
+#if defined(USE_OPENMP)
+      blas_unlock(&memory[position].lock);      
+    }
+#endif
    position ++;

  } while (position < NUM_BUFFERS);
+#if defined(SMP) && !defined(USE_OPENMP)
  UNLOCK_COMMAND(&alloc_lock);	
-
+#endif
  goto error;

  allocation :
@ -2645,10 +2730,11 @@ void *blas_memory_alloc(int procpos){
 #endif

  memory[position].used = 1;
-
+#if defined(SMP) && !defined(USE_OPENMP)
  UNLOCK_COMMAND(&alloc_lock);
-/*  blas_unlock(&memory[position].lock);*/
-
+#else
+  blas_unlock(&memory[position].lock);	
+#endif
  if (!memory[position].addr) {
    do {
 #ifdef DEBUG
@ -2693,9 +2779,13 @@ void *blas_memory_alloc(int procpos){

    } while ((BLASLONG)map_address == -1);

+#if defined(SMP) && !defined(USE_OPENMP)
    LOCK_COMMAND(&alloc_lock);
+#endif    
    memory[position].addr = map_address;
+#if defined(SMP) && !defined(USE_OPENMP)
    UNLOCK_COMMAND(&alloc_lock);
+#endif

 #ifdef DEBUG
    printf("  Mapping Succeeded. %p(%d)\n", (void *)memory[position].addr, position);
@ -2749,8 +2839,9 @@ void blas_memory_free(void *free_area){
 #endif

  position = 0;
+#if defined(SMP) && !defined(USE_OPENMP)
  LOCK_COMMAND(&alloc_lock);
-
+#endif
  while ((position < NUM_BUFFERS) && (memory[position].addr != free_area))
    position++;

@ -2764,7 +2855,9 @@ void blas_memory_free(void *free_area){
  WMB;

  memory[position].used = 0;
+#if defined(SMP) && !defined(USE_OPENMP)
  UNLOCK_COMMAND(&alloc_lock);
+#endif

 #ifdef DEBUG
  printf("Unmap Succeeded.\n\n");
@ -2779,8 +2872,9 @@ void blas_memory_free(void *free_area){
  for (position = 0; position < NUM_BUFFERS; position++)
    printf("%4ld  %p : %d\n", position, memory[position].addr, memory[position].used);
 #endif
+#if defined(SMP) && !defined(USE_OPENMP)
  UNLOCK_COMMAND(&alloc_lock);
-
+#endif
  return;
 }

--- a/exports/Makefile
+++ b/exports/Makefile
@ -141,6 +141,14 @@ else
 	$(OBJCOPY) --redefine-syms objcopy.def ../$(LIBNAME) ../$(LIBNAME).renamed
 ../$(LIBSONAME) : ../$(LIBNAME).renamed linktest.c
 endif
+
+ifeq ($(F_COMPILER), INTEL)
+	$(FC) $(FFLAGS) $(LDFLAGS) -shared -o ../$(LIBSONAME) \
+	-Wl,--whole-archive $< -Wl,--no-whole-archive \
+	-Wl,-soname,$(INTERNALNAME) $(EXTRALIB)
+	$(CC) $(CFLAGS) $(LDFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) $(FEXTRALIB) && echo OK.
+else
+
 ifneq ($(C_COMPILER), LSB)
 	$(CC) $(CFLAGS) $(LDFLAGS) -shared -o ../$(LIBSONAME) \
 	-Wl,--whole-archive $< -Wl,--no-whole-archive \
@ -152,6 +160,7 @@ else
 	-Wl,--whole-archive $< -Wl,--no-whole-archive \
 	-Wl,-soname,$(INTERNALNAME) $(EXTRALIB)
 	$(FC) $(CFLAGS) $(LDFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) $(FEXTRALIB) && echo OK.
+endif
 endif
 	rm -f linktest

--- a/exports/dllinit.c
+++ b/exports/dllinit.c
@ -40,15 +40,25 @@

 void gotoblas_init(void);
 void gotoblas_quit(void);
+#if defined(SMP) && defined(USE_TLS)
+void blas_thread_memory_cleanup(void);
+#endif

 BOOL APIENTRY DllMain(HINSTANCE hInst, DWORD reason, LPVOID reserved) {
-
-  if (reason == DLL_PROCESS_ATTACH) {
-    gotoblas_init();
-  }
-
-  if (reason == DLL_PROCESS_DETACH) {
-    gotoblas_quit();
+  switch(reason) {
+      case DLL_PROCESS_ATTACH:
+        gotoblas_init();
+        break;
+      case DLL_PROCESS_DETACH:
+        gotoblas_quit();
+        break;
+      case DLL_THREAD_ATTACH:
+        break;
+      case DLL_THREAD_DETACH:
+#if defined(SMP) && defined(USE_TLS)
+        blas_thread_memory_cleanup();
+#endif
+        break;
  }

  return TRUE;
--- a/getarch.c
+++ b/getarch.c
@ -91,6 +91,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include <unistd.h>
 #endif

+#if (( defined(__GNUC__)  && __GNUC__   > 6 && defined(__AVX2__)) || (defined(__clang__) && __clang_major__ >= 6))
+#else
+#define NO_AVX512
+#endif
 /* #define FORCE_P2		*/
 /* #define FORCE_KATMAI		*/
 /* #define FORCE_COPPERMINE	*/
@ -327,6 +331,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif

 #ifdef FORCE_SKYLAKEX
+#ifdef NO_AVX512
+#define FORCE
+#define FORCE_INTEL
+#define ARCHITECTURE    "X86"
+#define SUBARCHITECTURE "HASWELL"
+#define ARCHCONFIG   "-DHASWELL " \
+		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
+		     "-DL2_SIZE=262144 -DL2_LINESIZE=64 " \
+		     "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+		     "-DHAVE_CMOV -DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSSE3 -DHAVE_SSE4_1 -DHAVE_SSE4_2 -DHAVE_AVX " \
+                     "-DFMA3"
+#define LIBNAME   "haswell"
+#define CORENAME  "HASWELL"
+#else
 #define FORCE
 #define FORCE_INTEL
 #define ARCHITECTURE    "X86"
@ -340,6 +358,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define LIBNAME   "skylakex"
 #define CORENAME  "SKYLAKEX"
 #endif
+#endif

 #ifdef FORCE_ATOM
 #define FORCE
@ -1058,6 +1077,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #else
 #endif

+#ifdef FORCE_TSV110
+#define FORCE
+#define ARCHITECTURE    "ARM64"
+#define SUBARCHITECTURE "TSV110"
+#define SUBDIRNAME      "arm64"
+#define ARCHCONFIG   "-DTSV110 " \
+       "-DL1_CODE_SIZE=65536  -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=4 " \
+       "-DL1_DATA_SIZE=65536  -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=4 " \
+       "-DL2_SIZE=524288 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=8 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8"
+#define LIBNAME   "tsv110"
+#define CORENAME  "TSV110"
+#else
+#endif
+
+
 #ifdef FORCE_ZARCH_GENERIC
 #define FORCE
 #define ARCHITECTURE    "ZARCH"
@ -1078,6 +1114,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CORENAME  "Z13"
 #endif

+#ifdef FORCE_Z14
+#define FORCE
+#define ARCHITECTURE    "ZARCH"
+#define SUBARCHITECTURE "Z14"
+#define ARCHCONFIG   "-DZ14 " \
+       "-DDTB_DEFAULT_ENTRIES=64"
+#define LIBNAME   "z14"
+#define CORENAME  "Z14"
+#endif
+
 #ifndef FORCE

 #ifdef USER_TARGET
--- a/interface/trmv.c
+++ b/interface/trmv.c
@ -218,10 +218,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
  buffer = (FLOAT *)blas_memory_alloc(1);

 #ifdef SMP
-/*  nthreads = num_cpu_avail(2);
-
-FIXME trmv_thread was found to be broken, see issue 1332 */
-  nthreads = 1;
+  nthreads = num_cpu_avail(2);

  if (nthreads == 1) {
 #endif
--- a/interface/trsm.c
+++ b/interface/trsm.c
@ -81,6 +81,12 @@
 #endif
 #endif

+#ifndef COMPLEX
+#define SMP_FACTOR 256
+#else
+#define SMP_FACTOR 128
+#endif
+
 static int (*trsm[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) = {
 #ifndef TRMM
  TRSM_LNUU, TRSM_LNUN, TRSM_LNLU, TRSM_LNLN,
@ -366,11 +372,15 @@ void CNAME(enum CBLAS_ORDER order,
  mode |= (trans << BLAS_TRANSA_SHIFT);
  mode |= (side  << BLAS_RSIDE_SHIFT);

-  if ( args.m < 2*GEMM_MULTITHREAD_THRESHOLD )
+/*
+  if ( args.m < 2 * GEMM_MULTITHREAD_THRESHOLD )
 	args.nthreads = 1;
  else
-	if ( args.n < 2*GEMM_MULTITHREAD_THRESHOLD )
+	if ( args.n < 2 * GEMM_MULTITHREAD_THRESHOLD )
 		args.nthreads = 1;
+*/
+  if ( args.m * args.n < SMP_FACTOR * GEMM_MULTITHREAD_THRESHOLD)
+	args.nthreads = 1;
  else
 	args.nthreads = num_cpu_avail(3);
 		
--- a/interface/ztrmv.c
+++ b/interface/ztrmv.c
@ -239,9 +239,6 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
  } else
      nthreads = 1;

-/* FIXME TRMV multithreading appears to be broken, see issue 1332*/
-  nthreads = 1;
-
  if(nthreads > 1) {
    buffer_size = n > 16 ? 0 : n * 4 + 40;
  }
--- a/kernel/Makefile.L3
+++ b/kernel/Makefile.L3
@ -24,7 +24,7 @@ ifeq ($(TARGET), LOONGSON3B)
 USE_TRMM = 1
 endif

-ifeq ($(TARGET), GENERIC)
+ifeq ($(CORE), GENERIC)
 USE_TRMM = 1
 endif

@ -52,6 +52,10 @@ ifeq ($(ARCH), zarch)
 USE_TRMM = 1
 endif

+ifeq ($(CORE), Z14)
+USE_TRMM = 1
+endif
+



--- a/kernel/arm/imin.c
+++ b/kernel/arm/imin.c
@ -53,7 +53,7 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)

 	while(i < n)
 	{
-		if( x[ix] > minf )
+		if( x[ix] < minf )
 		{
 			min = i;
 			minf = x[ix];
--- a/kernel/arm64/KERNEL.TSV110
+++ b/kernel/arm64/KERNEL.TSV110
@ -0,0 +1,175 @@
+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
+
+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
+
+STRMMKERNEL	= ../generic/trmmkernel_4x4.c
+DTRMMKERNEL	= ../generic/trmmkernel_2x2.c
+CTRMMKERNEL	= ../generic/ztrmmkernel_2x2.c
+ZTRMMKERNEL	= ../generic/ztrmmkernel_2x2.c
+
+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
+
+SAMAXKERNEL  = amax.S
+DAMAXKERNEL  = amax.S
+CAMAXKERNEL  = zamax.S
+ZAMAXKERNEL  = zamax.S
+
+ISAMAXKERNEL = iamax.S
+IDAMAXKERNEL = iamax.S
+ICAMAXKERNEL = izamax.S
+IZAMAXKERNEL = izamax.S
+
+SASUMKERNEL  = asum.S
+DASUMKERNEL  = asum.S
+CASUMKERNEL  = casum.S
+ZASUMKERNEL  = zasum.S
+
+SAXPYKERNEL  = axpy.S
+DAXPYKERNEL  = axpy.S
+CAXPYKERNEL  = zaxpy.S
+ZAXPYKERNEL  = zaxpy.S
+
+SCOPYKERNEL  = copy.S
+DCOPYKERNEL  = copy.S
+CCOPYKERNEL  = copy.S
+ZCOPYKERNEL  = copy.S
+
+SDOTKERNEL   = dot.S
+DDOTKERNEL   = dot.S
+CDOTKERNEL   = zdot.S
+ZDOTKERNEL   = zdot.S
+DSDOTKERNEL  = dot.S
+
+SNRM2KERNEL  = nrm2.S
+DNRM2KERNEL  = nrm2.S
+CNRM2KERNEL  = znrm2.S
+ZNRM2KERNEL  = znrm2.S
+
+SROTKERNEL   = rot.S
+DROTKERNEL   = rot.S
+CROTKERNEL   = zrot.S
+ZROTKERNEL   = zrot.S
+
+SSCALKERNEL  = scal.S
+DSCALKERNEL  = scal.S
+CSCALKERNEL  = zscal.S
+ZSCALKERNEL  = zscal.S
+
+SSWAPKERNEL  = swap.S
+DSWAPKERNEL  = swap.S
+CSWAPKERNEL  = swap.S
+ZSWAPKERNEL  = swap.S
+
+SGEMVNKERNEL = gemv_n.S
+DGEMVNKERNEL = gemv_n.S
+CGEMVNKERNEL = zgemv_n.S
+ZGEMVNKERNEL = zgemv_n.S
+
+SGEMVTKERNEL = gemv_t.S
+DGEMVTKERNEL = gemv_t.S
+CGEMVTKERNEL = zgemv_t.S
+ZGEMVTKERNEL = zgemv_t.S
+
+SGEMMKERNEL    =  sgemm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N).S
+STRMMKERNEL    =  strmm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N).S
+ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N))
+SGEMMINCOPY    =  ../generic/gemm_ncopy_$(SGEMM_UNROLL_M).c
+SGEMMITCOPY    =  ../generic/gemm_tcopy_$(SGEMM_UNROLL_M).c
+SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
+SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+SGEMMONCOPY    =  ../generic/gemm_ncopy_$(SGEMM_UNROLL_N).c
+SGEMMOTCOPY    =  ../generic/gemm_tcopy_$(SGEMM_UNROLL_N).c
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+DGEMMKERNEL    =  dgemm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N).S
+DTRMMKERNEL    =  dtrmm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N).S
+
+ifneq ($(DGEMM_UNROLL_M), $(DGEMM_UNROLL_N))
+
+ifeq ($(DGEMM_UNROLL_M), 8)
+DGEMMINCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_M).S
+DGEMMITCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_M).S
+else
+DGEMMINCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_M).c
+DGEMMITCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_M).c
+endif
+
+DGEMMINCOPYOBJ =  dgemm_incopy$(TSUFFIX).$(SUFFIX)
+DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+ifeq ($(DGEMM_UNROLL_N), 4)
+DGEMMONCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_N).S
+DGEMMOTCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_N).S
+else
+DGEMMONCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_N).c
+DGEMMOTCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_N).c
+endif
+
+DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+CGEMMKERNEL    =  cgemm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N).S
+CTRMMKERNEL    =  ctrmm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N).S
+ifneq ($(CGEMM_UNROLL_M), $(CGEMM_UNROLL_N))
+CGEMMINCOPY    =  ../generic/zgemm_ncopy_$(CGEMM_UNROLL_M).c
+CGEMMITCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_M).c
+CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+CGEMMONCOPY    =  ../generic/zgemm_ncopy_$(CGEMM_UNROLL_N).c
+CGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_N).c
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+ZGEMMKERNEL    =  zgemm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
+ZTRMMKERNEL    =  ztrmm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
+ifneq ($(ZGEMM_UNROLL_M), $(ZGEMM_UNROLL_N))
+ZGEMMINCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_M).c
+ZGEMMITCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_M).c
+ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
+ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+ZGEMMONCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_N).c
+ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_N).c
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
--- a/kernel/mips/imin.c
+++ b/kernel/mips/imin.c
@ -45,7 +45,7 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)

 	while(i < n)
 	{
-		if( x[ix] > minf )
+		if( x[ix] < minf )
 		{
 			min = i;
 			minf = x[ix];
--- a/kernel/power/gemm_beta.S
+++ b/kernel/power/gemm_beta.S
@ -129,7 +129,7 @@ LL(12):
 	STFD	f0,  14 * SIZE(CO1)
 	STFD	f0,  15 * SIZE(CO1)

-	dcbst	PRE, CO1
+	dcbtst	PRE, CO1
 	addi	CO1, CO1,  16 * SIZE
 	bdnz	LL(12)
 	.align 4
--- a/kernel/power/zgemm_beta.S
+++ b/kernel/power/zgemm_beta.S
@ -134,7 +134,7 @@ LL(12):
 	STFD	f0,  14 * SIZE(CO1)
 	STFD	f0,  15 * SIZE(CO1)

-	dcbst	PRE, CO1
+	dcbtst	PRE, CO1
 	addi	CO1, CO1,  16 * SIZE
 	bdnz	LL(12)
 	.align 4
--- a/kernel/x86_64/caxpy_microk_bulldozer-2.c
+++ b/kernel/x86_64/caxpy_microk_bulldozer-2.c
@ -114,9 +114,9 @@ static void caxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vzeroupper					    \n\t"

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha),  // 4
@ -181,9 +181,9 @@ static void caxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vzeroupper					    \n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha),  // 4
--- a/kernel/x86_64/caxpy_microk_haswell-2.c
+++ b/kernel/x86_64/caxpy_microk_haswell-2.c
@ -112,9 +112,9 @@ static void caxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vzeroupper					    \n\t"

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha),  // 4
--- a/kernel/x86_64/caxpy_microk_sandy-2.c
+++ b/kernel/x86_64/caxpy_microk_sandy-2.c
@ -96,9 +96,9 @@ static void caxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vzeroupper					    \n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha),  // 4
--- a/kernel/x86_64/caxpy_microk_steamroller-2.c
+++ b/kernel/x86_64/caxpy_microk_steamroller-2.c
@ -114,9 +114,9 @@ static void caxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vzeroupper					    \n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha),  // 4
@ -181,9 +181,9 @@ static void caxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vzeroupper					    \n\t"

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha),  // 4
--- a/kernel/x86_64/cdot_microk_bulldozer-2.c
+++ b/kernel/x86_64/cdot_microk_bulldozer-2.c
@ -97,9 +97,9 @@ static void cdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vzeroupper		        	\n\t"

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
@ -176,9 +176,9 @@ static void cdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vzeroupper		        	\n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/cdot_microk_haswell-2.c
+++ b/kernel/x86_64/cdot_microk_haswell-2.c
@ -98,9 +98,9 @@ static void cdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vzeroupper					     \n\t"

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/cdot_microk_sandy-2.c
+++ b/kernel/x86_64/cdot_microk_sandy-2.c
@ -106,9 +106,9 @@ static void cdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vzeroupper					     \n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/cdot_microk_steamroller-2.c
+++ b/kernel/x86_64/cdot_microk_steamroller-2.c
@ -97,9 +97,9 @@ static void cdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vzeroupper		        	\n\t"

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
@ -176,9 +176,9 @@ static void cdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vzeroupper		        	\n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/cscal_microk_bulldozer-2.c
+++ b/kernel/x86_64/cscal_microk_bulldozer-2.c
@ -116,11 +116,11 @@ static void cscal_kernel_16( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	:
+	  "+r" (n),  	// 0
+          "+r" (x)      // 1
        :
-	  "r" (n),  	// 0
-          "r" (x),      // 1
          "r" (alpha)   // 2
-	: "cc", //"%0", "%1",
+	: "cc",
 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
@ -208,11 +208,11 @@ static void cscal_kernel_16_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	:
+	  "+r" (n),  	// 0
+          "+r" (x)      // 1
        :
-	  "r" (n),  	// 0
-          "r" (x),      // 1
          "r" (alpha)   // 2
-	: "cc", //"%0", "%1",
+	: "cc",
 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
@ -285,11 +285,11 @@ static void cscal_kernel_16_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	:
+	  "+r" (n),  	// 0
+          "+r" (x)      // 1
        :
-	  "r" (n),  	// 0
-          "r" (x),      // 1
          "r" (alpha)   // 2
-	: "cc", //"%0", "%1",
+	: "cc",
 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
@ -330,11 +330,11 @@ static void cscal_kernel_16_zero( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	:
+	  "+r" (n),  	// 0
+          "+r" (x)      // 1
        :
-	  "r" (n),  	// 0
-          "r" (x),      // 1
          "r" (alpha)   // 2
-	: "cc", //"%0", "%1",
+	: "cc",
 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
--- a/kernel/x86_64/cscal_microk_haswell-2.c
+++ b/kernel/x86_64/cscal_microk_haswell-2.c
@ -116,11 +116,11 @@ static void cscal_kernel_16( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	:
+	  "+r" (n),  	// 0
+          "+r" (x)      // 1
        :
-	  "r" (n),  	// 0
-          "r" (x),      // 1
          "r" (alpha)   // 2
-	: "cc", //"0", "1",
+	: "cc",
 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
@ -208,9 +208,9 @@ static void cscal_kernel_16_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	:
+	  "+r" (n),  	// 0
+          "+r" (x)      // 1
        :
-	  "r" (n),  	// 0
-          "r" (x),      // 1
          "r" (alpha)   // 2
 	: "cc", // "0", "1",
 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
@ -285,9 +285,9 @@ static void cscal_kernel_16_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	:
+	  "+r" (n),  	// 0
+          "+r" (x)      // 1
        :
-	  "r" (n),  	// 0
-          "r" (x),      // 1
          "r" (alpha)   // 2
 	: "cc", //"%0", "%1",
 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
@ -330,11 +330,11 @@ static void cscal_kernel_16_zero( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	: 
+	  "+r" (n),  	// 0
+          "+r" (x)      // 1
        :
-	  "r" (n),  	// 0
-          "r" (x),      // 1
          "r" (alpha)   // 2
-	: "cc", //"0", "1",
+	: "cc",
 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
--- a/kernel/x86_64/cscal_microk_steamroller-2.c
+++ b/kernel/x86_64/cscal_microk_steamroller-2.c
@ -117,11 +117,11 @@ static void cscal_kernel_16( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	:
-        : 
-	  "r" (n),  	// 0
-          "r" (x),      // 1
+	  "+r" (n),  	// 0
+          "+r" (x)      // 1
+	:
          "r" (alpha)   // 2
-	: "cc", //"0", "1",
+	: "cc",
 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
@ -209,11 +209,11 @@ static void cscal_kernel_16_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	: 
-        : 
-	  "r" (n),  	// 0
-          "r" (x),      // 1
+	  "+r" (n),  	// 0
+          "+r" (x)      // 1
+	:
          "r" (alpha)   // 2
-	: "cc", //"0", "1",
+	: "cc",
 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
@ -286,11 +286,11 @@ static void cscal_kernel_16_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	:
-        : 
-	  "r" (n),  	// 0
-          "r" (x),      // 1
+	  "+r" (n),  	// 0
+          "+r" (x)      // 1
+	:
          "r" (alpha)   // 2
-	: "cc", //"%0", "%1",
+	: "cc",
 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
@ -331,11 +331,11 @@ static void cscal_kernel_16_zero( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	:
-        : 
-	  "r" (n),  	// 0
-          "r" (x),      // 1
+	  "+r" (n),  	// 0
+          "+r" (x)      // 1
+	:
          "r" (alpha)   // 2
-	: "cc", //"0", "1",
+	: "cc",
 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
--- a/kernel/x86_64/daxpy_microk_bulldozer-2.c
+++ b/kernel/x86_64/daxpy_microk_bulldozer-2.c
@ -64,9 +64,9 @@ static void daxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"jnz		1b		             \n\t"

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha)   // 4
--- a/kernel/x86_64/daxpy_microk_haswell-2.c
+++ b/kernel/x86_64/daxpy_microk_haswell-2.c
@ -60,9 +60,9 @@ static void daxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vzeroupper				     \n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha)   // 4
--- a/kernel/x86_64/daxpy_microk_nehalem-2.c
+++ b/kernel/x86_64/daxpy_microk_nehalem-2.c
@ -73,9 +73,9 @@ static void daxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"jnz		1b		             \n\t"

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha)   // 4
--- a/kernel/x86_64/daxpy_microk_piledriver-2.c
+++ b/kernel/x86_64/daxpy_microk_piledriver-2.c
@ -79,9 +79,9 @@ static void daxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"jnz		1b		             			\n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha)   // 4
@ -141,9 +141,9 @@ static void daxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"jnz		1b		             			\n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha)   // 4
--- a/kernel/x86_64/daxpy_microk_sandy-2.c
+++ b/kernel/x86_64/daxpy_microk_sandy-2.c
@ -100,9 +100,9 @@ static void daxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vzeroupper					     \n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha)   // 4
--- a/kernel/x86_64/daxpy_microk_steamroller-2.c
+++ b/kernel/x86_64/daxpy_microk_steamroller-2.c
@ -79,9 +79,9 @@ static void daxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"jnz		1b		             			\n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha)   // 4
@ -141,9 +141,9 @@ static void daxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"jnz		1b		             			\n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha)   // 4
--- a/kernel/x86_64/ddot_microk_bulldozer-2.c
+++ b/kernel/x86_64/ddot_microk_bulldozer-2.c
@ -66,9 +66,9 @@ static void ddot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vmovsd		%%xmm4,    (%4)		\n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/ddot_microk_haswell-2.c
+++ b/kernel/x86_64/ddot_microk_haswell-2.c
@ -77,9 +77,9 @@ static void ddot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vzeroupper				\n\t"

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/ddot_microk_nehalem-2.c
+++ b/kernel/x86_64/ddot_microk_nehalem-2.c
@ -76,9 +76,9 @@ static void ddot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"movsd	       %%xmm4,    (%4)	\n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/ddot_microk_piledriver-2.c
+++ b/kernel/x86_64/ddot_microk_piledriver-2.c
@ -82,9 +82,9 @@ static void ddot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vzeroupper					     \n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
@ -146,9 +146,9 @@ static void ddot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vzeroupper					     \n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/ddot_microk_sandy-2.c
+++ b/kernel/x86_64/ddot_microk_sandy-2.c
@ -82,9 +82,9 @@ static void ddot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vzeroupper				\n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/ddot_microk_steamroller-2.c
+++ b/kernel/x86_64/ddot_microk_steamroller-2.c
@ -79,9 +79,9 @@ static void ddot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vzeroupper					     \n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/dgemv_n_4.c
+++ b/kernel/x86_64/dgemv_n_4.c
@ -111,9 +111,9 @@ static void dgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 	"jnz		1b		       \n\t"

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (ap[0]),  // 4
@ -166,9 +166,9 @@ static void dgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *a
        "jnz            1b               \n\t"

        :
+          "+r" (i),     // 0    
+          "+r" (n)      // 1
        :
-          "r" (i),      // 0    
-          "r" (n),      // 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (ap),     // 4
--- a/kernel/x86_64/dgemv_n_microk_haswell-4.c
+++ b/kernel/x86_64/dgemv_n_microk_haswell-4.c
@ -104,6 +104,7 @@ static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
          "r" (ap[3]),  // 7
          "r" (alpha)   // 8
 	: "cc", 
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3",
 	  "%xmm4", "%xmm5", 
 	  "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", 
--- a/kernel/x86_64/dgemv_n_microk_piledriver-4.c
+++ b/kernel/x86_64/dgemv_n_microk_piledriver-4.c
@ -38,42 +38,42 @@ static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	__asm__  __volatile__
 	(
 	"vzeroupper			 \n\t"
-	"vbroadcastsd    (%2), %%ymm12	 \n\t"	// x0 
-	"vbroadcastsd   8(%2), %%ymm13	 \n\t"	// x1 
-	"vbroadcastsd  16(%2), %%ymm14	 \n\t"	// x2 
-	"vbroadcastsd  24(%2), %%ymm15	 \n\t"	// x3 
-	"vbroadcastsd  32(%2), %%ymm0 	 \n\t"	// x4 
-	"vbroadcastsd  40(%2), %%ymm1 	 \n\t"	// x5 
-	"vbroadcastsd  48(%2), %%ymm2 	 \n\t"	// x6 
-	"vbroadcastsd  56(%2), %%ymm3 	 \n\t"	// x7 
+	"vbroadcastsd    (%3), %%ymm12	 \n\t"	// x0 
+	"vbroadcastsd   8(%3), %%ymm13	 \n\t"	// x1 
+	"vbroadcastsd  16(%3), %%ymm14	 \n\t"	// x2 
+	"vbroadcastsd  24(%3), %%ymm15	 \n\t"	// x3 
+	"vbroadcastsd  32(%3), %%ymm0 	 \n\t"	// x4 
+	"vbroadcastsd  40(%3), %%ymm1 	 \n\t"	// x5 
+	"vbroadcastsd  48(%3), %%ymm2 	 \n\t"	// x6 
+	"vbroadcastsd  56(%3), %%ymm3 	 \n\t"	// x7 

 	"vbroadcastsd    (%9), %%ymm6 	 \n\t"	// alpha 

        "testq          $0x04, %1                      \n\t"
        "jz             2f                     \n\t"

-	"vmovupd	(%3,%0,8), %%ymm7	       \n\t"	// 4 * y
+	"vmovupd	(%4,%0,8), %%ymm7	       \n\t"	// 4 * y
 	"vxorpd		%%ymm4 , %%ymm4, %%ymm4        \n\t"
 	"vxorpd		%%ymm5 , %%ymm5, %%ymm5        \n\t"

-	"vfmadd231pd   (%4,%0,8), %%ymm12, %%ymm4      \n\t" 
-	"vfmadd231pd   (%5,%0,8), %%ymm13, %%ymm5      \n\t" 
-	"vfmadd231pd   (%6,%0,8), %%ymm14, %%ymm4      \n\t" 
-	"vfmadd231pd   (%7,%0,8), %%ymm15, %%ymm5      \n\t" 
+	"vfmadd231pd   (%5,%0,8), %%ymm12, %%ymm4      \n\t" 
+	"vfmadd231pd   (%6,%0,8), %%ymm13, %%ymm5      \n\t" 
+	"vfmadd231pd   (%7,%0,8), %%ymm14, %%ymm4      \n\t" 
+	"vfmadd231pd   (%8,%0,8), %%ymm15, %%ymm5      \n\t" 

-	"vfmadd231pd   (%4,%8,8), %%ymm0 , %%ymm4      \n\t" 
-	"vfmadd231pd   (%5,%8,8), %%ymm1 , %%ymm5      \n\t" 
-	"vfmadd231pd   (%6,%8,8), %%ymm2 , %%ymm4      \n\t" 
-	"vfmadd231pd   (%7,%8,8), %%ymm3 , %%ymm5      \n\t" 
+	"vfmadd231pd   (%5,%2,8), %%ymm0 , %%ymm4      \n\t" 
+	"vfmadd231pd   (%6,%2,8), %%ymm1 , %%ymm5      \n\t" 
+	"vfmadd231pd   (%7,%2,8), %%ymm2 , %%ymm4      \n\t" 
+	"vfmadd231pd   (%8,%2,8), %%ymm3 , %%ymm5      \n\t" 

 	"vaddpd		%%ymm4 , %%ymm5 , %%ymm5       \n\t"
 	"vmulpd		%%ymm6 , %%ymm5 , %%ymm5       \n\t"
 	"vaddpd		%%ymm7 , %%ymm5 , %%ymm5       \n\t"


-	"vmovupd  %%ymm5,   (%3,%0,8)		       \n\t"	// 4 * y
+	"vmovupd  %%ymm5,   (%4,%0,8)		       \n\t"	// 4 * y

-        "addq		$4 , %8	  	 	       \n\t"
+        "addq		$4 , %2	  	 	       \n\t"
        "addq		$4 , %0	  	 	       \n\t"
 	"subq	        $4 , %1			       \n\t"		

@ -88,35 +88,35 @@ static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO

 	"vxorpd		%%ymm4 , %%ymm4, %%ymm4        \n\t"
 	"vxorpd		%%ymm5 , %%ymm5, %%ymm5        \n\t"
-	"vmovupd	(%3,%0,8), %%ymm8	       \n\t"	// 4 * y
-	"vmovupd      32(%3,%0,8), %%ymm9	       \n\t"	// 4 * y
+	"vmovupd	(%4,%0,8), %%ymm8	       \n\t"	// 4 * y
+	"vmovupd      32(%4,%0,8), %%ymm9	       \n\t"	// 4 * y

-	"vfmadd231pd   (%4,%0,8), %%ymm12, %%ymm4      \n\t" 
-	"vfmadd231pd 32(%4,%0,8), %%ymm12, %%ymm5      \n\t" 
-	"vfmadd231pd   (%5,%0,8), %%ymm13, %%ymm4      \n\t" 
-	"vfmadd231pd 32(%5,%0,8), %%ymm13, %%ymm5      \n\t" 
-	"vfmadd231pd   (%6,%0,8), %%ymm14, %%ymm4      \n\t" 
-	"vfmadd231pd 32(%6,%0,8), %%ymm14, %%ymm5      \n\t" 
-	"vfmadd231pd   (%7,%0,8), %%ymm15, %%ymm4      \n\t" 
-	"vfmadd231pd 32(%7,%0,8), %%ymm15, %%ymm5      \n\t" 
+	"vfmadd231pd   (%5,%0,8), %%ymm12, %%ymm4      \n\t" 
+	"vfmadd231pd 32(%5,%0,8), %%ymm12, %%ymm5      \n\t" 
+	"vfmadd231pd   (%6,%0,8), %%ymm13, %%ymm4      \n\t" 
+	"vfmadd231pd 32(%6,%0,8), %%ymm13, %%ymm5      \n\t" 
+	"vfmadd231pd   (%7,%0,8), %%ymm14, %%ymm4      \n\t" 
+	"vfmadd231pd 32(%7,%0,8), %%ymm14, %%ymm5      \n\t" 
+	"vfmadd231pd   (%8,%0,8), %%ymm15, %%ymm4      \n\t" 
+	"vfmadd231pd 32(%8,%0,8), %%ymm15, %%ymm5      \n\t" 

-	"vfmadd231pd   (%4,%8,8), %%ymm0 , %%ymm4      \n\t" 
+	"vfmadd231pd   (%5,%2,8), %%ymm0 , %%ymm4      \n\t" 
        "addq		$8 , %0	  	 	       \n\t"
-	"vfmadd231pd 32(%4,%8,8), %%ymm0 , %%ymm5      \n\t" 
-	"vfmadd231pd   (%5,%8,8), %%ymm1 , %%ymm4      \n\t" 
-	"vfmadd231pd 32(%5,%8,8), %%ymm1 , %%ymm5      \n\t" 
-	"vfmadd231pd   (%6,%8,8), %%ymm2 , %%ymm4      \n\t" 
-	"vfmadd231pd 32(%6,%8,8), %%ymm2 , %%ymm5      \n\t" 
-	"vfmadd231pd   (%7,%8,8), %%ymm3 , %%ymm4      \n\t" 
-	"vfmadd231pd 32(%7,%8,8), %%ymm3 , %%ymm5      \n\t" 
+	"vfmadd231pd 32(%5,%2,8), %%ymm0 , %%ymm5      \n\t" 
+	"vfmadd231pd   (%6,%2,8), %%ymm1 , %%ymm4      \n\t" 
+	"vfmadd231pd 32(%6,%2,8), %%ymm1 , %%ymm5      \n\t" 
+	"vfmadd231pd   (%7,%2,8), %%ymm2 , %%ymm4      \n\t" 
+	"vfmadd231pd 32(%7,%2,8), %%ymm2 , %%ymm5      \n\t" 
+	"vfmadd231pd   (%8,%2,8), %%ymm3 , %%ymm4      \n\t" 
+	"vfmadd231pd 32(%8,%2,8), %%ymm3 , %%ymm5      \n\t" 

 	"vfmadd231pd     %%ymm6 , %%ymm4 , %%ymm8      \n\t"
 	"vfmadd231pd     %%ymm6 , %%ymm5 , %%ymm9      \n\t"

-        "addq		$8 , %8	  	 	      \n\t"
+        "addq		$8 , %2	  	 	      \n\t"
 	"vmovupd  %%ymm8,-64(%3,%0,8)		      \n\t"	// 4 * y
 	"subq	        $8 , %1			      \n\t"		
-	"vmovupd  %%ymm9,-32(%3,%0,8)		      \n\t"	// 4 * y
+	"vmovupd  %%ymm9,-32(%4,%0,8)		      \n\t"	// 4 * y

 	"jnz		1b		      \n\t"

@ -125,15 +125,15 @@ static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO

 	:
          "+r" (i),	// 0	
-	  "+r" (n)  	// 1
+	  "+r" (n),  	// 1
+          "+r" (lda4)   // 2
        : 
-          "r" (x),      // 2
-          "r" (y),      // 3
-          "r" (ap[0]),  // 4
-          "r" (ap[1]),  // 5
-          "r" (ap[2]),  // 6
-          "r" (ap[3]),  // 7
-          "r" (lda4),   // 8
+          "r" (x),      // 3
+          "r" (y),      // 4
+          "r" (ap[0]),  // 5
+          "r" (ap[1]),  // 6
+          "r" (ap[2]),  // 7
+          "r" (ap[3]),  // 8
          "r" (alpha)   // 9
 	: "cc", 
 	  "%xmm0", "%xmm1", 
--- a/kernel/x86_64/dgemv_t_4.c
+++ b/kernel/x86_64/dgemv_t_4.c
@ -127,9 +127,9 @@ static void dgemv_kernel_4x2(BLASLONG n, FLOAT *ap0, FLOAT *ap1, FLOAT *x, FLOAT
 	"movsd	       %%xmm11,8(%2)	        \n\t"

        :
-   	:
-	"r" (i),	 // 0
-	"r" (n),	 // 1
+	"+r" (i),	 // 0
+	"+r" (n)	 // 1
+	:
        "r" (y),         // 2    
        "r" (ap0),       // 3
        "r" (ap1),       // 4
@ -195,9 +195,9 @@ static void dgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y)
 	"movsd	       %%xmm10, (%2)	        \n\t"

        :
-   	:
-	"r" (i),	 // 0
-	"r" (n),	 // 1
+	"+r" (i),	 // 0
+	"+r" (n)	 // 1
+	:
        "r" (y),         // 2    
        "r" (ap),        // 3
        "r" (x)          // 4
@ -259,9 +259,9 @@ static void add_y(BLASLONG n, FLOAT da , FLOAT *src, FLOAT *dest, BLASLONG inc_d
        "jnz            1b              \n\t"

        :
-   	:
-	"r" (i),	  // 0
-	"r" (n),	  // 1
+	"+r" (i),	  // 0
+	"+r" (n)	  // 1
+	:
        "r" (&da),        // 2    
        "r" (src),        // 3
        "r" (dest)        // 4
--- a/kernel/x86_64/dger_microk_sandy-2.c
+++ b/kernel/x86_64/dger_microk_sandy-2.c
@ -105,9 +105,9 @@ static void dger_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vzeroupper					     \n\t"

 	:
-        : 
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
+	:
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha)   // 4
--- a/kernel/x86_64/dscal.c
+++ b/kernel/x86_64/dscal.c
@ -136,8 +136,8 @@ static void dscal_kernel_inc_8(BLASLONG n, FLOAT *alpha, FLOAT *x, BLASLONG inc_
 	"jnz    1b					    \n\t"

        :
+          "+r" (n)      // 0
        :
-          "r" (n),      // 0
          "r" (x),      // 1
          "r" (x1),     // 2
          "r" (alpha),  // 3
--- a/kernel/x86_64/dscal_microk_bulldozer-2.c
+++ b/kernel/x86_64/dscal_microk_bulldozer-2.c
@ -122,9 +122,9 @@ static void dscal_kernel_8( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	:
-        : 
-	  "r" (n1),  	// 0
-          "r" (x),      // 1
+	  "+r" (n1),  	// 0
+          "+r" (x)      // 1
+	:
          "r" (alpha),  // 2
 	  "r" (n2)   	// 3
 	: "cc", 
@ -188,9 +188,9 @@ static void dscal_kernel_8_zero( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	:
-        : 
-	  "r" (n1),  	// 0
-          "r" (x),      // 1
+	  "+r" (n1),  	// 0
+          "+r" (x)      // 1
+	:
          "r" (alpha),  // 2
 	  "r" (n2)   	// 3
 	: "cc", 
--- a/kernel/x86_64/dscal_microk_haswell-2.c
+++ b/kernel/x86_64/dscal_microk_haswell-2.c
@ -122,9 +122,9 @@ static void dscal_kernel_8( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	:
-        : 
-	  "r" (n1),  	// 0
-          "r" (x),      // 1
+	  "+r" (n1),  	// 0
+          "+r" (x)      // 1
+	:
          "r" (alpha),  // 2
 	  "r" (n2)   	// 3
 	: "cc", 
@ -188,9 +188,9 @@ static void dscal_kernel_8_zero( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	: 
-        : 
-	  "r" (n1),  	// 0
-          "r" (x),      // 1
+	  "+r" (n1),  	// 0
+          "+r" (x)      // 1
+	:
          "r" (alpha),  // 2
 	  "r" (n2)   	// 3
 	: "cc", 
--- a/kernel/x86_64/dscal_microk_sandy-2.c
+++ b/kernel/x86_64/dscal_microk_sandy-2.c
@ -122,9 +122,9 @@ static void dscal_kernel_8( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	:
-        : 
-	  "r" (n1),  	// 0
-          "r" (x),      // 1
+	  "+r" (n1),  	// 0
+          "+r" (x)      // 1
+	:
          "r" (alpha),  // 2
 	  "r" (n2)   	// 3
 	: "cc", 
@ -188,9 +188,9 @@ static void dscal_kernel_8_zero( BLASLONG n, FLOAT *alpha, FLOAT *x)
 	"vzeroupper					    \n\t"

 	: 
-        : 
-	  "r" (n1),  	// 0
-          "r" (x),      // 1
+	  "+r" (n1),  	// 0
+          "+r" (x)      // 1
+	:
          "r" (alpha),  // 2
 	  "r" (n2)   	// 3
 	: "cc", 
--- a/kernel/x86_64/dsymv_L_microk_bulldozer-2.c
+++ b/kernel/x86_64/dsymv_L_microk_bulldozer-2.c
@ -113,8 +113,8 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
 	"vmovsd         %%xmm3 ,24(%9)		\n\t"	// save temp2

 	:
+          "+r" (from)	// 0	
        :
-          "r" (from),	// 0	
 	  "r" (to),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
--- a/kernel/x86_64/dsymv_L_microk_haswell-2.c
+++ b/kernel/x86_64/dsymv_L_microk_haswell-2.c
@ -105,8 +105,8 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
 	"vzeroupper				     \n\t"

 	:
+          "+r" (from)	// 0	
        :
-          "r" (from),	// 0	
 	  "r" (to),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
--- a/kernel/x86_64/dsymv_L_microk_nehalem-2.c
+++ b/kernel/x86_64/dsymv_L_microk_nehalem-2.c
@ -108,8 +108,8 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
 	"movsd         %%xmm3 , 24(%9)		\n\t"	// save temp2

 	:
+          "+r" (from)	// 0	
        :
-          "r" (from),	// 0	
 	  "r" (to),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
--- a/kernel/x86_64/dsymv_L_microk_sandy-2.c
+++ b/kernel/x86_64/dsymv_L_microk_sandy-2.c
@ -114,8 +114,8 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
 	"vzeroupper				     \n\t"

 	:
+          "+r" (from)	// 0	
        :
-          "r" (from),	// 0	
 	  "r" (to),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
--- a/kernel/x86_64/dsymv_U_microk_bulldozer-2.c
+++ b/kernel/x86_64/dsymv_U_microk_bulldozer-2.c
@ -106,9 +106,9 @@ static void dsymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
 	"vmovsd         %%xmm3 ,24(%9)		\n\t"	// save temp2

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (a0),     // 4
--- a/kernel/x86_64/dsymv_U_microk_haswell-2.c
+++ b/kernel/x86_64/dsymv_U_microk_haswell-2.c
@ -107,9 +107,9 @@ static void dsymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
 	"vzeroupper				     \n\t"

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (a0),	// 4
--- a/kernel/x86_64/dsymv_U_microk_nehalem-2.c
+++ b/kernel/x86_64/dsymv_U_microk_nehalem-2.c
@ -101,9 +101,9 @@ static void dsymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
 	"movsd         %%xmm3 , 24(%9)		\n\t"	// save temp2

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (a0),     // 4
--- a/kernel/x86_64/dsymv_U_microk_sandy-2.c
+++ b/kernel/x86_64/dsymv_U_microk_sandy-2.c
@ -116,9 +116,9 @@ static void dsymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
 	"vzeroupper				     \n\t"

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (a0),	// 4
--- a/kernel/x86_64/dtrsm_kernel_RN_haswell.c
+++ b/kernel/x86_64/dtrsm_kernel_RN_haswell.c
@ -119,9 +119,9 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	cmpq	       $0, %0						\n\t"
 	"	je	       4f						\n\t"

-	"	vmovups         (%2,%1,4), %%ymm0				\n\t"	// read a
-	"	vmovups         (%3,%1,8), %%ymm1				\n\t"	// read b0
-	"	vmovups       32(%3,%1,8), %%ymm2				\n\t"	// read b1
+	"	vmovups         (%8,%1,4), %%ymm0				\n\t"	// read a
+	"	vmovups         (%9,%1,8), %%ymm1				\n\t"	// read b0
+	"	vmovups       32(%9,%1,8), %%ymm2				\n\t"	// read b1


 	"	addq		$8, %1						\n\t"
@ -131,18 +131,18 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	.p2align 4							\n\t"
 	"1:									\n\t"

-	"	vmovups         (%2,%1,4), %%ymm4				\n\t"	// read a
+	"	vmovups         (%8,%1,4), %%ymm4				\n\t"	// read a
        "       vpermpd         $0xb1  , %%ymm0 , %%ymm3                	\n\t"

 	"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm8			\n\t"
 	"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm12			\n\t"

-	"	vmovups         (%3,%1,8), %%ymm5				\n\t"	// read b0
+	"	vmovups         (%9,%1,8), %%ymm5				\n\t"	// read b0
 	"	vfmadd231pd	%%ymm3 , %%ymm1 , %%ymm9			\n\t"
 	"	vfmadd231pd	%%ymm3 , %%ymm2 , %%ymm13			\n\t"

        "       vpermpd         $0x1b  , %%ymm3 , %%ymm0                	\n\t"
-	"	vmovups       32(%3,%1,8), %%ymm6				\n\t"	// read b1
+	"	vmovups       32(%9,%1,8), %%ymm6				\n\t"	// read b1
        "       vpermpd         $0xb1  , %%ymm0 , %%ymm3                	\n\t"
 	"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm10			\n\t"
 	"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm14			\n\t"
@ -155,18 +155,18 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON

 	"	jz		22f						\n\t"

-	"	vmovups         (%2,%1,4), %%ymm0				\n\t"	// read a
+	"	vmovups         (%8,%1,4), %%ymm0				\n\t"	// read a

 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm8			\n\t"
 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm12			\n\t"

        "       vpermpd         $0xb1  , %%ymm4 , %%ymm4                	\n\t"
-	"	vmovups         (%3,%1,8), %%ymm1				\n\t"	// read b0
+	"	vmovups         (%9,%1,8), %%ymm1				\n\t"	// read b0
 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm9			\n\t"
 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm13			\n\t"

        "       vpermpd         $0x1b  , %%ymm4 , %%ymm4                	\n\t"
-	"	vmovups       32(%3,%1,8), %%ymm2				\n\t"	// read b1
+	"	vmovups       32(%9,%1,8), %%ymm2				\n\t"	// read b1
 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm10			\n\t"
 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm14			\n\t"

@ -268,7 +268,7 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vmovups		  (%6,%7,1) , %%ymm7			\n\t"	// read c7

 	"	vsubpd		%%ymm8 , %%ymm0 , %%ymm8		\n\t"
-	"	vmovups           (%9),  %%ymm0				\n\t"
+	"	vmovups           (%3),  %%ymm0				\n\t"
 	"	vsubpd		%%ymm9 , %%ymm1 , %%ymm9		\n\t"
 	"	vpermpd		$0x55 ,  %%ymm0 , %%ymm1		\n\t"
 	"	vsubpd		%%ymm10, %%ymm2 , %%ymm10		\n\t"
@ -278,7 +278,7 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vpermpd		$0x00 ,  %%ymm0 , %%ymm0		\n\t"

 	"	vsubpd		%%ymm12, %%ymm4 , %%ymm12		\n\t"
-	"	vmovups         32(%9),  %%ymm4				\n\t"
+	"	vmovups         32(%3),  %%ymm4				\n\t"
 	"	vsubpd		%%ymm13, %%ymm5 , %%ymm13		\n\t"
 	"	vpermpd		$0x55 ,  %%ymm4 , %%ymm5		\n\t"
 	"	vsubpd		%%ymm14, %%ymm6 , %%ymm14		\n\t"
@ -290,15 +290,15 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON

 	"5:								\n\t"	// i = 0

-	"	addq	$64, %9						\n\t"	// b=b+8
+	"	addq	$64, %3						\n\t"	// b=b+8

 	"	vmulpd		%%ymm8 , %%ymm0, %%ymm8			\n\t"	// a *bb
-	"	vmovups           (%9),  %%ymm0				\n\t"
-	"	vmovups		%%ymm8 , (%8)				\n\t"	// write a
+	"	vmovups           (%3),  %%ymm0				\n\t"
+	"	vmovups		%%ymm8 , (%2)				\n\t"	// write a
 	"	vmovups		%%ymm8 , (%4)				\n\t"	// write c

 	"	vfnmadd231pd	%%ymm8 , %%ymm1 , %%ymm9		\n\t"
-	"	vmovups         32(%9),  %%ymm1				\n\t"
+	"	vmovups         32(%3),  %%ymm1				\n\t"
 	"	vfnmadd231pd	%%ymm8 , %%ymm2 , %%ymm10		\n\t"
 	"	vpermpd		$0xaa ,  %%ymm0 , %%ymm2		\n\t"
 	"	vfnmadd231pd	%%ymm8 , %%ymm3 , %%ymm11		\n\t"
@ -313,15 +313,15 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vpermpd		$0xff ,  %%ymm1 , %%ymm7		\n\t"
 	"	vpermpd		$0x00 ,  %%ymm1 , %%ymm4		\n\t"

-	"	addq	$64, %9						\n\t"	// b=b+8
-	"	addq	$32, %8						\n\t"	// a=a+8
+	"	addq	$64, %3						\n\t"	// b=b+8
+	"	addq	$32, %2						\n\t"	// a=a+8



 	"	vmulpd		%%ymm9 , %%ymm0, %%ymm9			\n\t"	// a *bb
-	"	vmovups           (%9),  %%ymm0				\n\t"
-	"	vmovups         32(%9),  %%ymm1				\n\t"
-	"	vmovups		%%ymm9 , (%8)				\n\t"	// write a
+	"	vmovups           (%3),  %%ymm0				\n\t"
+	"	vmovups         32(%3),  %%ymm1				\n\t"
+	"	vmovups		%%ymm9 , (%2)				\n\t"	// write a
 	"	vmovups		%%ymm9 , (%4,%7,1)			\n\t"	// write c

 	"	vfnmadd231pd	%%ymm9 , %%ymm2 , %%ymm10		\n\t"
@ -337,13 +337,13 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vpermpd		$0xff ,  %%ymm1 , %%ymm7		\n\t"
 	"	vpermpd		$0x00 ,  %%ymm1 , %%ymm4		\n\t"

-	"	addq	$64, %9						\n\t"	// b=b+8
-	"	addq	$32, %8						\n\t"	// a=a+8
+	"	addq	$64, %3						\n\t"	// b=b+8
+	"	addq	$32, %2						\n\t"	// a=a+8

 	"	vmulpd		%%ymm10, %%ymm0, %%ymm10		\n\t"	// a *bb
-	"	vmovups           (%9),  %%ymm0				\n\t"
-	"	vmovups         32(%9),  %%ymm1				\n\t"
-	"	vmovups		%%ymm10, (%8)				\n\t"	// write a
+	"	vmovups           (%3),  %%ymm0				\n\t"
+	"	vmovups         32(%3),  %%ymm1				\n\t"
+	"	vmovups		%%ymm10, (%2)				\n\t"	// write a
 	"	vmovups		%%ymm10, (%4,%7,2)			\n\t"	// write c

 	"	vfnmadd231pd	%%ymm10, %%ymm3 , %%ymm11		\n\t"
@ -358,14 +358,14 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vpermpd		$0x00 ,  %%ymm1 , %%ymm4		\n\t"


-	"	addq	$64, %9						\n\t"	// b=b+8
-	"	addq	$32, %8						\n\t"	// a=a+8
+	"	addq	$64, %3						\n\t"	// b=b+8
+	"	addq	$32, %2						\n\t"	// a=a+8



 	"	vmulpd		%%ymm11, %%ymm0, %%ymm11		\n\t"	// a *bb
-	"	vmovups         32(%9),  %%ymm1				\n\t"
-	"	vmovups		%%ymm11, (%8)				\n\t"	// write a
+	"	vmovups         32(%3),  %%ymm1				\n\t"
+	"	vmovups		%%ymm11, (%2)				\n\t"	// write a
 	"	vmovups		%%ymm11, (%5)     			\n\t"	// write c

 	"	vfnmadd231pd	%%ymm11, %%ymm4 , %%ymm12		\n\t"
@ -378,13 +378,13 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vpermpd		$0x00 ,  %%ymm1 , %%ymm0		\n\t"


-	"	addq	$64, %9						\n\t"	// b=b+8
-	"	addq	$32, %8						\n\t"	// a=a+8
+	"	addq	$64, %3						\n\t"	// b=b+8
+	"	addq	$32, %2						\n\t"	// a=a+8


 	"	vmulpd		%%ymm12, %%ymm0, %%ymm12		\n\t"	// a *bb
-	"	vmovups         32(%9),  %%ymm1				\n\t"
-	"	vmovups		%%ymm12, (%8)				\n\t"	// write a
+	"	vmovups         32(%3),  %%ymm1				\n\t"
+	"	vmovups		%%ymm12, (%2)				\n\t"	// write a
 	"	vmovups		%%ymm12, (%5,%7,1)			\n\t"	// write c

 	"	vfnmadd231pd	%%ymm12, %%ymm5 , %%ymm13		\n\t"
@ -394,12 +394,12 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vpermpd		$0xff ,  %%ymm1 , %%ymm7		\n\t"
 	"	vpermpd		$0x55 ,  %%ymm1 , %%ymm0		\n\t"

-	"	addq	$64, %9						\n\t"	// b=b+8
-	"	addq	$32, %8						\n\t"	// a=a+8
+	"	addq	$64, %3						\n\t"	// b=b+8
+	"	addq	$32, %2						\n\t"	// a=a+8

 	"	vmulpd		%%ymm13, %%ymm0, %%ymm13		\n\t"	// a *bb
-	"	vmovups         32(%9),  %%ymm1				\n\t"
-	"	vmovups		%%ymm13, (%8)				\n\t"	// write a
+	"	vmovups         32(%3),  %%ymm1				\n\t"
+	"	vmovups		%%ymm13, (%2)				\n\t"	// write a
 	"	vmovups		%%ymm13, (%5,%7,2)			\n\t"	// write c

 	"	vfnmadd231pd	%%ymm13, %%ymm6 , %%ymm14		\n\t"
@ -408,39 +408,39 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vpermpd		$0xaa ,  %%ymm1 , %%ymm0		\n\t"


-	"	addq	$64, %9						\n\t"	// b=b+8
-	"	addq	$32, %8						\n\t"	// a=a+8
+	"	addq	$64, %3						\n\t"	// b=b+8
+	"	addq	$32, %2						\n\t"	// a=a+8


 	"	vmulpd		%%ymm14, %%ymm0, %%ymm14		\n\t"	// a *bb
-	"	vmovups         32(%9),  %%ymm1				\n\t"
-	"	vmovups		%%ymm14, (%8)				\n\t"	// write a
+	"	vmovups         32(%3),  %%ymm1				\n\t"
+	"	vmovups		%%ymm14, (%2)				\n\t"	// write a
 	"	vmovups		%%ymm14, (%6)     			\n\t"	// write c

 	"	vfnmadd231pd	%%ymm14, %%ymm7 , %%ymm15		\n\t"

 	"	vpermpd		$0xff ,  %%ymm1 , %%ymm0		\n\t"

-	"	addq	$32, %8						\n\t"	// a=a+8
+	"	addq	$32, %2						\n\t"	// a=a+8

 	"	vmulpd		%%ymm15, %%ymm0, %%ymm15		\n\t"	// a *bb
-	"	vmovups		%%ymm15, (%8)				\n\t"	// write a
+	"	vmovups		%%ymm15, (%2)				\n\t"	// write a
 	"	vmovups		%%ymm15, (%6,%7,1)			\n\t"	// write c

 	"	vzeroupper						\n\t"

        :
+          "+r" (n1),     // 0    
+          "+a" (i),      // 1    
+          "+r" (as),     // 2
+          "+r" (bs)      // 3
        :
-          "r" (n1),     // 0    
-          "a" (i),      // 1    
-          "r" (a),      // 2
-          "r" (b),      // 3
          "r" (c),      // 4
          "r" (c3),     // 5
          "r" (c6),     // 6
          "r" (ldc),    // 7
-          "r" (as),     // 8
-          "r" (bs)      // 9
+          "r" (a),     // 8
+          "r" (b)      // 9
        : "cc",
          "%xmm0", "%xmm1", "%xmm2", "%xmm3",
          "%xmm4", "%xmm5", "%xmm6", "%xmm7",
--- a/kernel/x86_64/dtrsm_kernel_RT_bulldozer.c
+++ b/kernel/x86_64/dtrsm_kernel_RT_bulldozer.c
@ -125,14 +125,14 @@ static void dtrsm_RT_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	.align 16							\n\t"
 	"1:									\n\t"

-	"	prefetcht0	384(%2,%1,8)					\n\t"
-	"	prefetcht0	384(%3,%1,8)					\n\t"
-	"	vmovddup	(%3,%1,2), %%xmm0				\n\t"	// read b
-	"	vmovups         (%2,%1,8), %%xmm4				\n\t"
-	"	vmovddup       8(%3,%1,2), %%xmm1				\n\t"	
-	"	vmovups       16(%2,%1,8), %%xmm5				\n\t"
-	"	vmovups       32(%2,%1,8), %%xmm6				\n\t"
-	"	vmovups       48(%2,%1,8), %%xmm7				\n\t"
+	"	prefetcht0	384(%6,%1,8)					\n\t"
+	"	prefetcht0	384(%7,%1,8)					\n\t"
+	"	vmovddup	(%7,%1,2), %%xmm0				\n\t"	// read b
+	"	vmovups         (%6,%1,8), %%xmm4				\n\t"
+	"	vmovddup       8(%7,%1,2), %%xmm1				\n\t"	
+	"	vmovups       16(%6,%1,8), %%xmm5				\n\t"
+	"	vmovups       32(%6,%1,8), %%xmm6				\n\t"
+	"	vmovups       48(%6,%1,8), %%xmm7				\n\t"

 	"	vfmaddpd	%%xmm8 , %%xmm0 , %%xmm4 , %%xmm8		\n\t"
 	"	vfmaddpd	%%xmm12, %%xmm1 , %%xmm4 , %%xmm12		\n\t"
@ -147,13 +147,13 @@ static void dtrsm_RT_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON

 	"	jz		2f						\n\t"

-	"	prefetcht0	384(%2,%1,8)					\n\t"
-	"	vmovddup	(%3,%1,2), %%xmm0				\n\t"	// read b
-	"	vmovups         (%2,%1,8), %%xmm4				\n\t"
-	"	vmovddup       8(%3,%1,2), %%xmm1				\n\t"	
-	"	vmovups       16(%2,%1,8), %%xmm5				\n\t"
-	"	vmovups       32(%2,%1,8), %%xmm6				\n\t"
-	"	vmovups       48(%2,%1,8), %%xmm7				\n\t"
+	"	prefetcht0	384(%6,%1,8)					\n\t"
+	"	vmovddup	(%7,%1,2), %%xmm0				\n\t"	// read b
+	"	vmovups         (%6,%1,8), %%xmm4				\n\t"
+	"	vmovddup       8(%7,%1,2), %%xmm1				\n\t"	
+	"	vmovups       16(%6,%1,8), %%xmm5				\n\t"
+	"	vmovups       32(%6,%1,8), %%xmm6				\n\t"
+	"	vmovups       48(%6,%1,8), %%xmm7				\n\t"

 	"	vfmaddpd	%%xmm8 , %%xmm0 , %%xmm4 , %%xmm8		\n\t"
 	"	vfmaddpd	%%xmm12, %%xmm1 , %%xmm4 , %%xmm12		\n\t"
@ -168,13 +168,13 @@ static void dtrsm_RT_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON

 	"	jz		2f						\n\t"

-	"	prefetcht0	384(%2,%1,8)					\n\t"
-	"	vmovddup	(%3,%1,2), %%xmm0				\n\t"	// read b
-	"	vmovups         (%2,%1,8), %%xmm4				\n\t"
-	"	vmovddup       8(%3,%1,2), %%xmm1				\n\t"	
-	"	vmovups       16(%2,%1,8), %%xmm5				\n\t"
-	"	vmovups       32(%2,%1,8), %%xmm6				\n\t"
-	"	vmovups       48(%2,%1,8), %%xmm7				\n\t"
+	"	prefetcht0	384(%6,%1,8)					\n\t"
+	"	vmovddup	(%7,%1,2), %%xmm0				\n\t"	// read b
+	"	vmovups         (%6,%1,8), %%xmm4				\n\t"
+	"	vmovddup       8(%7,%1,2), %%xmm1				\n\t"	
+	"	vmovups       16(%6,%1,8), %%xmm5				\n\t"
+	"	vmovups       32(%6,%1,8), %%xmm6				\n\t"
+	"	vmovups       48(%6,%1,8), %%xmm7				\n\t"

 	"	vfmaddpd	%%xmm8 , %%xmm0 , %%xmm4 , %%xmm8		\n\t"
 	"	vfmaddpd	%%xmm12, %%xmm1 , %%xmm4 , %%xmm12		\n\t"
@ -189,13 +189,13 @@ static void dtrsm_RT_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON

 	"	jz		2f						\n\t"

-	"	prefetcht0	384(%2,%1,8)					\n\t"
-	"	vmovddup	(%3,%1,2), %%xmm0				\n\t"	// read b
-	"	vmovddup       8(%3,%1,2), %%xmm1				\n\t"	
-	"	vmovups         (%2,%1,8), %%xmm4				\n\t"
-	"	vmovups       16(%2,%1,8), %%xmm5				\n\t"
-	"	vmovups       32(%2,%1,8), %%xmm6				\n\t"
-	"	vmovups       48(%2,%1,8), %%xmm7				\n\t"
+	"	prefetcht0	384(%6,%1,8)					\n\t"
+	"	vmovddup	(%7,%1,2), %%xmm0				\n\t"	// read b
+	"	vmovddup       8(%7,%1,2), %%xmm1				\n\t"	
+	"	vmovups         (%6,%1,8), %%xmm4				\n\t"
+	"	vmovups       16(%6,%1,8), %%xmm5				\n\t"
+	"	vmovups       32(%6,%1,8), %%xmm6				\n\t"
+	"	vmovups       48(%6,%1,8), %%xmm7				\n\t"

 	"	vfmaddpd	%%xmm8 , %%xmm0 , %%xmm4 , %%xmm8		\n\t"
 	"	vfmaddpd	%%xmm12, %%xmm1 , %%xmm4 , %%xmm12		\n\t"
@ -235,18 +235,18 @@ static void dtrsm_RT_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON

 	"3:									\n\t"	// i = 1

-	"	vmovddup	(%7), %%xmm1					\n\t"	// read b
-	"	vmovddup       8(%7), %%xmm0					\n\t"	// read bb
+	"	vmovddup	(%3), %%xmm1					\n\t"	// read b
+	"	vmovddup       8(%3), %%xmm0					\n\t"	// read bb

 	"	vmulpd		%%xmm12 ,  %%xmm0 ,  %%xmm12			\n\t"	// aa * bb 
 	"	vmulpd		%%xmm13 ,  %%xmm0 ,  %%xmm13			\n\t"	// aa * bb 
 	"	vmulpd		%%xmm14 ,  %%xmm0 ,  %%xmm14			\n\t"	// aa * bb 
 	"	vmulpd		%%xmm15 ,  %%xmm0 ,  %%xmm15			\n\t"	// aa * bb 

-	"	vmovups		%%xmm12 ,    (%6)				\n\t"	// write a
-	"	vmovups		%%xmm13 ,  16(%6)				\n\t"	// write a
-	"	vmovups		%%xmm14 ,  32(%6)				\n\t"	// write a
-	"	vmovups		%%xmm15 ,  48(%6)				\n\t"	// write a
+	"	vmovups		%%xmm12 ,    (%2)				\n\t"	// write a
+	"	vmovups		%%xmm13 ,  16(%2)				\n\t"	// write a
+	"	vmovups		%%xmm14 ,  32(%2)				\n\t"	// write a
+	"	vmovups		%%xmm15 ,  48(%2)				\n\t"	// write a

 	"	vmovups		%%xmm12 ,    (%5)				\n\t"	// write c1
 	"	vmovups		%%xmm13 ,  16(%5)				\n\t"	
@ -259,20 +259,20 @@ static void dtrsm_RT_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vfnmaddpd	%%xmm11 ,  %%xmm15 , %%xmm1 , %%xmm11		\n\t"   

 	"									\n\t" // i = 0
-	"	subq		$16 , %7					\n\t" // b = b - 2
-	"	subq		$64 , %6					\n\t" // a = a - 8
+	"	subq		$16 , %3					\n\t" // b = b - 2
+	"	subq		$64 , %2					\n\t" // a = a - 8

-	"	vmovddup        (%7), %%xmm0					\n\t"	// read bb
+	"	vmovddup        (%3), %%xmm0					\n\t"	// read bb

 	"	vmulpd		%%xmm8  ,  %%xmm0 ,  %%xmm8 			\n\t"	// aa * bb 
 	"	vmulpd		%%xmm9  ,  %%xmm0 ,  %%xmm9 			\n\t"
 	"	vmulpd		%%xmm10 ,  %%xmm0 ,  %%xmm10			\n\t"
 	"	vmulpd		%%xmm11 ,  %%xmm0 ,  %%xmm11			\n\t"

-	"	vmovups		%%xmm8  ,    (%6)				\n\t"	// write a
-	"	vmovups		%%xmm9  ,  16(%6)				\n\t"
-	"	vmovups		%%xmm10 ,  32(%6)				\n\t"
-	"	vmovups		%%xmm11 ,  48(%6)				\n\t"
+	"	vmovups		%%xmm8  ,    (%2)				\n\t"	// write a
+	"	vmovups		%%xmm9  ,  16(%2)				\n\t"
+	"	vmovups		%%xmm10 ,  32(%2)				\n\t"
+	"	vmovups		%%xmm11 ,  48(%2)				\n\t"

 	"	vmovups		%%xmm8  ,    (%4)				\n\t"	// write c0
 	"	vmovups		%%xmm9  ,  16(%4)				\n\t"
@ -282,15 +282,15 @@ static void dtrsm_RT_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vzeroupper							\n\t"

        :
+          "+r" (n1),     // 0    
+          "+a" (i),      // 1    
+          "+r" (as),     // 2
+          "+r" (bs)      // 3
        :
-          "r" (n1),     // 0    
-          "a" (i),      // 1    
-          "r" (a),      // 2
-          "r" (b),      // 3
          "r" (c),      // 4
          "r" (c1),     // 5
-          "r" (as),     // 6
-          "r" (bs)      // 7
+          "r" (a),      // 6
+          "r" (b)       // 7
        : "cc",
          "%xmm0", "%xmm1", "%xmm2", "%xmm3",
          "%xmm4", "%xmm5", "%xmm6", "%xmm7",
--- a/kernel/x86_64/gemm_kernel_4x8_nano.S
+++ b/kernel/x86_64/gemm_kernel_4x8_nano.S
@ -135,7 +135,7 @@
 #endif

 	movq	%rsp, %rbx	# save old stack
-	subq	$128 + LOCAL_BUFFER_SIZE, %rsp
+	subq	$256 + LOCAL_BUFFER_SIZE, %rsp
 	andq	$-4096, %rsp	# align stack

 	STACK_TOUCHING
--- a/kernel/x86_64/gemm_kernel_8x4_sse.S
+++ b/kernel/x86_64/gemm_kernel_8x4_sse.S
@ -383,7 +383,7 @@
 	EMMS

 	movq	%rsp, %rbx	# save old stack
-	subq	$128 + LOCAL_BUFFER_SIZE, %rsp
+	subq	$256 + LOCAL_BUFFER_SIZE, %rsp
 	andq	$-4096, %rsp	# align stack

 	STACK_TOUCHING
--- a/kernel/x86_64/saxpy_microk_haswell-2.c
+++ b/kernel/x86_64/saxpy_microk_haswell-2.c
@ -60,9 +60,9 @@ static void saxpy_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vzeroupper				     \n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha)   // 4
--- a/kernel/x86_64/saxpy_microk_nehalem-2.c
+++ b/kernel/x86_64/saxpy_microk_nehalem-2.c
@ -73,9 +73,9 @@ static void saxpy_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"jnz		1b		             \n\t"

 	:
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha)   // 4
--- a/kernel/x86_64/saxpy_microk_piledriver-2.c
+++ b/kernel/x86_64/saxpy_microk_piledriver-2.c
@ -79,9 +79,9 @@ static void saxpy_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vzeroupper				     \n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha)   // 4
@ -140,9 +140,9 @@ static void saxpy_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vzeroupper				     \n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha)   // 4
--- a/kernel/x86_64/saxpy_microk_sandy-2.c
+++ b/kernel/x86_64/saxpy_microk_sandy-2.c
@ -100,9 +100,9 @@ static void saxpy_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vzeroupper					     \n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (alpha)   // 4
--- a/kernel/x86_64/sdot_microk_bulldozer-2.c
+++ b/kernel/x86_64/sdot_microk_bulldozer-2.c
@ -67,9 +67,9 @@ static void sdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vmovss		%%xmm4,    (%4)		\n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/sdot_microk_haswell-2.c
+++ b/kernel/x86_64/sdot_microk_haswell-2.c
@ -80,9 +80,9 @@ static void sdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vzeroupper				\n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/sdot_microk_nehalem-2.c
+++ b/kernel/x86_64/sdot_microk_nehalem-2.c
@ -76,9 +76,9 @@ static void sdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"movss	       %%xmm4,    (%4)	\n\t"

 	: 
+          "+r" (i),	// 0
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/sdot_microk_sandy-2.c
+++ b/kernel/x86_64/sdot_microk_sandy-2.c
@ -83,9 +83,9 @@ static void sdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vzeroupper				\n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/sdot_microk_steamroller-2.c
+++ b/kernel/x86_64/sdot_microk_steamroller-2.c
@ -81,9 +81,9 @@ static void sdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vmovss		%%xmm4,    (%4)		\n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
@ -144,9 +144,9 @@ static void sdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vmovss		%%xmm4,    (%4)		\n\t"

 	: 
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
        :
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (dot)     // 4
--- a/kernel/x86_64/sgemv_n_4.c
+++ b/kernel/x86_64/sgemv_n_4.c
@ -149,9 +149,9 @@ static void sgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 	"jnz		1b		       \n\t"

 	:
-        : 
-          "r" (i),	// 0	
-	  "r" (n),  	// 1
+          "+r" (i),	// 0	
+	  "+r" (n)  	// 1
+	:
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (ap[0]),  // 4
@ -223,9 +223,9 @@ static void sgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *a

        "3:      			 \n\t" 
        :
+          "+r" (i),     // 0    
+          "+r" (n1)     // 1
        :
-          "r" (i),      // 0    
-          "r" (n1),     // 1
          "r" (x),      // 2
          "r" (y),      // 3
          "r" (ap),     // 4
@ -277,9 +277,9 @@ static void add_y(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_dest)
        "jnz            1b              \n\t"

        :
+        "+r" (i),         // 0
+        "+r" (n)          // 1
        :
-        "r" (i),          // 0
-        "r" (n),          // 1
        "r" (src),        // 2
        "r" (dest)        // 3
        : "cc",
--- a/kernel/x86_64/sgemv_n_microk_bulldozer-4.c
+++ b/kernel/x86_64/sgemv_n_microk_bulldozer-4.c
@ -37,14 +37,14 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO

 	__asm__  __volatile__
 	(
-	"vbroadcastss    (%2), %%xmm12	 \n\t"	// x0 
-	"vbroadcastss   4(%2), %%xmm13	 \n\t"	// x1 
-	"vbroadcastss   8(%2), %%xmm14	 \n\t"	// x2 
-	"vbroadcastss  12(%2), %%xmm15	 \n\t"	// x3 
-	"vbroadcastss  16(%2), %%xmm0 	 \n\t"	// x4 
-	"vbroadcastss  20(%2), %%xmm1 	 \n\t"	// x5 
-	"vbroadcastss  24(%2), %%xmm2 	 \n\t"	// x6 
-	"vbroadcastss  28(%2), %%xmm3 	 \n\t"	// x7 
+	"vbroadcastss    (%3), %%xmm12	 \n\t"	// x0 
+	"vbroadcastss   4(%3), %%xmm13	 \n\t"	// x1 
+	"vbroadcastss   8(%3), %%xmm14	 \n\t"	// x2 
+	"vbroadcastss  12(%3), %%xmm15	 \n\t"	// x3 
+	"vbroadcastss  16(%3), %%xmm0 	 \n\t"	// x4 
+	"vbroadcastss  20(%3), %%xmm1 	 \n\t"	// x5 
+	"vbroadcastss  24(%3), %%xmm2 	 \n\t"	// x6 
+	"vbroadcastss  28(%3), %%xmm3 	 \n\t"	// x7 

 	"vbroadcastss    (%9), %%xmm8 	 \n\t"	// alpha 

@ -54,22 +54,22 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"vxorps		%%xmm4, %%xmm4 , %%xmm4  \n\t"
 	"vxorps		%%xmm5, %%xmm5 , %%xmm5  \n\t"

-	"vfmaddps %%xmm4,   (%4,%0,4), %%xmm12, %%xmm4 \n\t" 
-	"vfmaddps %%xmm5,   (%5,%0,4), %%xmm13, %%xmm5 \n\t" 
-	"vfmaddps %%xmm4,   (%6,%0,4), %%xmm14, %%xmm4 \n\t" 
-	"vfmaddps %%xmm5,   (%7,%0,4), %%xmm15, %%xmm5 \n\t" 
+	"vfmaddps %%xmm4,   (%5,%0,4), %%xmm12, %%xmm4 \n\t" 
+	"vfmaddps %%xmm5,   (%6,%0,4), %%xmm13, %%xmm5 \n\t" 
+	"vfmaddps %%xmm4,   (%7,%0,4), %%xmm14, %%xmm4 \n\t" 
+	"vfmaddps %%xmm5,   (%8,%0,4), %%xmm15, %%xmm5 \n\t" 
        "addq		$4 , %0	  	 	       \n\t"

-	"vfmaddps %%xmm4,   (%4,%8,4), %%xmm0 , %%xmm4 \n\t" 
-	"vfmaddps %%xmm5,   (%5,%8,4), %%xmm1 , %%xmm5 \n\t" 
-	"vfmaddps %%xmm4,   (%6,%8,4), %%xmm2 , %%xmm4 \n\t" 
-	"vfmaddps %%xmm5,   (%7,%8,4), %%xmm3 , %%xmm5 \n\t" 
-        "addq		$4 , %8	  	 	       \n\t"
+	"vfmaddps %%xmm4,   (%5,%2,4), %%xmm0 , %%xmm4 \n\t" 
+	"vfmaddps %%xmm5,   (%6,%2,4), %%xmm1 , %%xmm5 \n\t" 
+	"vfmaddps %%xmm4,   (%7,%2,4), %%xmm2 , %%xmm4 \n\t" 
+	"vfmaddps %%xmm5,   (%8,%2,4), %%xmm3 , %%xmm5 \n\t" 
+        "addq		$4 , %2	  	 	       \n\t"
 	
 	"vaddps		%%xmm5 , %%xmm4, %%xmm4        \n\t"
-	"vfmaddps -16(%3,%0,4) , %%xmm4, %%xmm8,%%xmm6 \n\t"
+	"vfmaddps -16(%4,%0,4) , %%xmm4, %%xmm8,%%xmm6 \n\t"
 	"subq	        $4 , %1			       \n\t"		
-	"vmovups  %%xmm6, -16(%3,%0,4)		       \n\t"	// 4 * y
+	"vmovups  %%xmm6, -16(%4,%0,4)		       \n\t"	// 4 * y

 	"2:                                  \n\t"

@ -79,31 +79,31 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"vxorps		%%xmm4, %%xmm4 , %%xmm4  \n\t"
 	"vxorps		%%xmm5, %%xmm5 , %%xmm5  \n\t"

-	"vfmaddps %%xmm4,   (%4,%0,4), %%xmm12, %%xmm4 \n\t" 
-	"vfmaddps %%xmm5, 16(%4,%0,4), %%xmm12, %%xmm5 \n\t" 
-	"vfmaddps %%xmm4,   (%5,%0,4), %%xmm13, %%xmm4 \n\t" 
-	"vfmaddps %%xmm5, 16(%5,%0,4), %%xmm13, %%xmm5 \n\t" 
-	"vfmaddps %%xmm4,   (%6,%0,4), %%xmm14, %%xmm4 \n\t" 
-	"vfmaddps %%xmm5, 16(%6,%0,4), %%xmm14, %%xmm5 \n\t" 
-	"vfmaddps %%xmm4,   (%7,%0,4), %%xmm15, %%xmm4 \n\t" 
-	"vfmaddps %%xmm5, 16(%7,%0,4), %%xmm15, %%xmm5 \n\t" 
+	"vfmaddps %%xmm4,   (%5,%0,4), %%xmm12, %%xmm4 \n\t" 
+	"vfmaddps %%xmm5, 16(%5,%0,4), %%xmm12, %%xmm5 \n\t" 
+	"vfmaddps %%xmm4,   (%6,%0,4), %%xmm13, %%xmm4 \n\t" 
+	"vfmaddps %%xmm5, 16(%6,%0,4), %%xmm13, %%xmm5 \n\t" 
+	"vfmaddps %%xmm4,   (%7,%0,4), %%xmm14, %%xmm4 \n\t" 
+	"vfmaddps %%xmm5, 16(%7,%0,4), %%xmm14, %%xmm5 \n\t" 
+	"vfmaddps %%xmm4,   (%8,%0,4), %%xmm15, %%xmm4 \n\t" 
+	"vfmaddps %%xmm5, 16(%8,%0,4), %%xmm15, %%xmm5 \n\t" 

-	"vfmaddps %%xmm4,   (%4,%8,4), %%xmm0 , %%xmm4 \n\t" 
-        "vfmaddps %%xmm5, 16(%4,%8,4), %%xmm0 , %%xmm5 \n\t" 
-	"vfmaddps %%xmm4,   (%5,%8,4), %%xmm1 , %%xmm4 \n\t" 
-	"vfmaddps %%xmm5, 16(%5,%8,4), %%xmm1 , %%xmm5 \n\t" 
-	"vfmaddps %%xmm4,   (%6,%8,4), %%xmm2 , %%xmm4 \n\t" 
-	"vfmaddps %%xmm5, 16(%6,%8,4), %%xmm2 , %%xmm5 \n\t" 
-	"vfmaddps %%xmm4,   (%7,%8,4), %%xmm3 , %%xmm4 \n\t" 
-	"vfmaddps %%xmm5, 16(%7,%8,4), %%xmm3 , %%xmm5 \n\t" 
+	"vfmaddps %%xmm4,   (%5,%2,4), %%xmm0 , %%xmm4 \n\t" 
+        "vfmaddps %%xmm5, 16(%5,%2,4), %%xmm0 , %%xmm5 \n\t" 
+	"vfmaddps %%xmm4,   (%6,%2,4), %%xmm1 , %%xmm4 \n\t" 
+	"vfmaddps %%xmm5, 16(%6,%2,4), %%xmm1 , %%xmm5 \n\t" 
+	"vfmaddps %%xmm4,   (%7,%2,4), %%xmm2 , %%xmm4 \n\t" 
+	"vfmaddps %%xmm5, 16(%7,%2,4), %%xmm2 , %%xmm5 \n\t" 
+	"vfmaddps %%xmm4,   (%8,%2,4), %%xmm3 , %%xmm4 \n\t" 
+	"vfmaddps %%xmm5, 16(%8,%2,4), %%xmm3 , %%xmm5 \n\t" 
 	
-	"vfmaddps    (%3,%0,4) , %%xmm4,%%xmm8,%%xmm4 \n\t"
-	"vfmaddps  16(%3,%0,4) , %%xmm5,%%xmm8,%%xmm5 \n\t"
-	"vmovups  %%xmm4,   (%3,%0,4)		      \n\t"	// 4 * y
-	"vmovups  %%xmm5, 16(%3,%0,4)		      \n\t"	// 4 * y
+	"vfmaddps    (%4,%0,4) , %%xmm4,%%xmm8,%%xmm4 \n\t"
+	"vfmaddps  16(%4,%0,4) , %%xmm5,%%xmm8,%%xmm5 \n\t"
+	"vmovups  %%xmm4,   (%4,%0,4)		      \n\t"	// 4 * y
+	"vmovups  %%xmm5, 16(%4,%0,4)		      \n\t"	// 4 * y

        "addq		$8 , %0	  	 	      \n\t"
-        "addq		$8 , %8	  	 	      \n\t"
+        "addq		$8 , %2	  	 	      \n\t"
 	"subq	        $8 , %1			      \n\t"		


@ -120,62 +120,62 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"vxorps		%%xmm6, %%xmm6 , %%xmm6  \n\t"
 	"vxorps		%%xmm7, %%xmm7 , %%xmm7  \n\t"

-        "prefetcht0      192(%4,%0,4)                  \n\t"
-	"vfmaddps %%xmm4,   (%4,%0,4), %%xmm12, %%xmm4 \n\t" 
-	"vfmaddps %%xmm5, 16(%4,%0,4), %%xmm12, %%xmm5 \n\t" 
        "prefetcht0      192(%5,%0,4)                  \n\t"
-	"vfmaddps %%xmm4,   (%5,%0,4), %%xmm13, %%xmm4 \n\t" 
-	"vfmaddps %%xmm5, 16(%5,%0,4), %%xmm13, %%xmm5 \n\t" 
+	"vfmaddps %%xmm4,   (%5,%0,4), %%xmm12, %%xmm4 \n\t" 
+	"vfmaddps %%xmm5, 16(%5,%0,4), %%xmm12, %%xmm5 \n\t" 
        "prefetcht0      192(%6,%0,4)                  \n\t"
-	"vfmaddps %%xmm4,   (%6,%0,4), %%xmm14, %%xmm4 \n\t" 
-	"vfmaddps %%xmm5, 16(%6,%0,4), %%xmm14, %%xmm5 \n\t" 
+	"vfmaddps %%xmm4,   (%6,%0,4), %%xmm13, %%xmm4 \n\t" 
+	"vfmaddps %%xmm5, 16(%6,%0,4), %%xmm13, %%xmm5 \n\t" 
        "prefetcht0      192(%7,%0,4)                  \n\t"
-	"vfmaddps %%xmm4,   (%7,%0,4), %%xmm15, %%xmm4 \n\t" 
+	"vfmaddps %%xmm4,   (%7,%0,4), %%xmm14, %%xmm4 \n\t" 
+	"vfmaddps %%xmm5, 16(%7,%0,4), %%xmm14, %%xmm5 \n\t" 
+        "prefetcht0      192(%8,%0,4)                  \n\t"
+	"vfmaddps %%xmm4,   (%8,%0,4), %%xmm15, %%xmm4 \n\t" 
 	".align 2				 \n\t"
-	"vfmaddps %%xmm5, 16(%7,%0,4), %%xmm15, %%xmm5 \n\t" 
+	"vfmaddps %%xmm5, 16(%8,%0,4), %%xmm15, %%xmm5 \n\t" 

-	"vfmaddps %%xmm6, 32(%4,%0,4), %%xmm12, %%xmm6 \n\t" 
-	"vfmaddps %%xmm7, 48(%4,%0,4), %%xmm12, %%xmm7 \n\t" 
-	"vfmaddps %%xmm6, 32(%5,%0,4), %%xmm13, %%xmm6 \n\t" 
-	"vfmaddps %%xmm7, 48(%5,%0,4), %%xmm13, %%xmm7 \n\t" 
-	"vfmaddps %%xmm6, 32(%6,%0,4), %%xmm14, %%xmm6 \n\t" 
-	"vfmaddps %%xmm7, 48(%6,%0,4), %%xmm14, %%xmm7 \n\t" 
-	"vfmaddps %%xmm6, 32(%7,%0,4), %%xmm15, %%xmm6 \n\t" 
-	"vfmaddps %%xmm7, 48(%7,%0,4), %%xmm15, %%xmm7 \n\t" 
+	"vfmaddps %%xmm6, 32(%5,%0,4), %%xmm12, %%xmm6 \n\t" 
+	"vfmaddps %%xmm7, 48(%5,%0,4), %%xmm12, %%xmm7 \n\t" 
+	"vfmaddps %%xmm6, 32(%6,%0,4), %%xmm13, %%xmm6 \n\t" 
+	"vfmaddps %%xmm7, 48(%6,%0,4), %%xmm13, %%xmm7 \n\t" 
+	"vfmaddps %%xmm6, 32(%7,%0,4), %%xmm14, %%xmm6 \n\t" 
+	"vfmaddps %%xmm7, 48(%7,%0,4), %%xmm14, %%xmm7 \n\t" 
+	"vfmaddps %%xmm6, 32(%8,%0,4), %%xmm15, %%xmm6 \n\t" 
+	"vfmaddps %%xmm7, 48(%8,%0,4), %%xmm15, %%xmm7 \n\t" 

-        "prefetcht0      192(%4,%8,4)                  \n\t"
-	"vfmaddps %%xmm4,   (%4,%8,4), %%xmm0 , %%xmm4 \n\t" 
-        "vfmaddps %%xmm5, 16(%4,%8,4), %%xmm0 , %%xmm5 \n\t" 
-        "prefetcht0      192(%5,%8,4)                  \n\t"
-	"vfmaddps %%xmm4,   (%5,%8,4), %%xmm1 , %%xmm4 \n\t" 
-	"vfmaddps %%xmm5, 16(%5,%8,4), %%xmm1 , %%xmm5 \n\t" 
-        "prefetcht0      192(%6,%8,4)                  \n\t"
-	"vfmaddps %%xmm4,   (%6,%8,4), %%xmm2 , %%xmm4 \n\t" 
-	"vfmaddps %%xmm5, 16(%6,%8,4), %%xmm2 , %%xmm5 \n\t" 
-        "prefetcht0      192(%7,%8,4)                  \n\t"
-	"vfmaddps %%xmm4,   (%7,%8,4), %%xmm3 , %%xmm4 \n\t" 
-	"vfmaddps %%xmm5, 16(%7,%8,4), %%xmm3 , %%xmm5 \n\t" 
+        "prefetcht0      192(%5,%2,4)                  \n\t"
+	"vfmaddps %%xmm4,   (%5,%2,4), %%xmm0 , %%xmm4 \n\t" 
+        "vfmaddps %%xmm5, 16(%5,%2,4), %%xmm0 , %%xmm5 \n\t" 
+        "prefetcht0      192(%6,%2,4)                  \n\t"
+	"vfmaddps %%xmm4,   (%6,%2,4), %%xmm1 , %%xmm4 \n\t" 
+	"vfmaddps %%xmm5, 16(%6,%2,4), %%xmm1 , %%xmm5 \n\t" 
+        "prefetcht0      192(%7,%2,4)                  \n\t"
+	"vfmaddps %%xmm4,   (%7,%2,4), %%xmm2 , %%xmm4 \n\t" 
+	"vfmaddps %%xmm5, 16(%7,%2,4), %%xmm2 , %%xmm5 \n\t" 
+        "prefetcht0      192(%8,%2,4)                  \n\t"
+	"vfmaddps %%xmm4,   (%8,%2,4), %%xmm3 , %%xmm4 \n\t" 
+	"vfmaddps %%xmm5, 16(%8,%2,4), %%xmm3 , %%xmm5 \n\t" 
 	
-	"vfmaddps %%xmm6, 32(%4,%8,4), %%xmm0 , %%xmm6 \n\t" 
-        "vfmaddps %%xmm7, 48(%4,%8,4), %%xmm0 , %%xmm7 \n\t" 
-	"vfmaddps %%xmm6, 32(%5,%8,4), %%xmm1 , %%xmm6 \n\t" 
-	"vfmaddps %%xmm7, 48(%5,%8,4), %%xmm1 , %%xmm7 \n\t" 
-	"vfmaddps %%xmm6, 32(%6,%8,4), %%xmm2 , %%xmm6 \n\t" 
-	"vfmaddps %%xmm7, 48(%6,%8,4), %%xmm2 , %%xmm7 \n\t" 
-	"vfmaddps %%xmm6, 32(%7,%8,4), %%xmm3 , %%xmm6 \n\t" 
-	"vfmaddps %%xmm7, 48(%7,%8,4), %%xmm3 , %%xmm7 \n\t" 
+	"vfmaddps %%xmm6, 32(%5,%2,4), %%xmm0 , %%xmm6 \n\t" 
+        "vfmaddps %%xmm7, 48(%5,%2,4), %%xmm0 , %%xmm7 \n\t" 
+	"vfmaddps %%xmm6, 32(%6,%2,4), %%xmm1 , %%xmm6 \n\t" 
+	"vfmaddps %%xmm7, 48(%6,%2,4), %%xmm1 , %%xmm7 \n\t" 
+	"vfmaddps %%xmm6, 32(%7,%2,4), %%xmm2 , %%xmm6 \n\t" 
+	"vfmaddps %%xmm7, 48(%7,%2,4), %%xmm2 , %%xmm7 \n\t" 
+	"vfmaddps %%xmm6, 32(%8,%2,4), %%xmm3 , %%xmm6 \n\t" 
+	"vfmaddps %%xmm7, 48(%8,%2,4), %%xmm3 , %%xmm7 \n\t" 
 	
-	"vfmaddps    (%3,%0,4) , %%xmm4,%%xmm8,%%xmm4 \n\t"
-	"vfmaddps  16(%3,%0,4) , %%xmm5,%%xmm8,%%xmm5 \n\t"
-	"vfmaddps  32(%3,%0,4) , %%xmm6,%%xmm8,%%xmm6 \n\t"
-	"vfmaddps  48(%3,%0,4) , %%xmm7,%%xmm8,%%xmm7 \n\t"
+	"vfmaddps    (%4,%0,4) , %%xmm4,%%xmm8,%%xmm4 \n\t"
+	"vfmaddps  16(%4,%0,4) , %%xmm5,%%xmm8,%%xmm5 \n\t"
+	"vfmaddps  32(%4,%0,4) , %%xmm6,%%xmm8,%%xmm6 \n\t"
+	"vfmaddps  48(%4,%0,4) , %%xmm7,%%xmm8,%%xmm7 \n\t"

        "addq		$16, %0	  	 	      \n\t"
-	"vmovups  %%xmm4,-64(%3,%0,4)		      \n\t"	// 4 * y
-	"vmovups  %%xmm5,-48(%3,%0,4)		      \n\t"	// 4 * y
-        "addq		$16, %8	  	 	      \n\t"
-	"vmovups  %%xmm6,-32(%3,%0,4)		      \n\t"	// 4 * y
-	"vmovups  %%xmm7,-16(%3,%0,4)		      \n\t"	// 4 * y
+	"vmovups  %%xmm4,-64(%4,%0,4)		      \n\t"	// 4 * y
+	"vmovups  %%xmm5,-48(%4,%0,4)		      \n\t"	// 4 * y
+        "addq		$16, %2	  	 	      \n\t"
+	"vmovups  %%xmm6,-32(%4,%0,4)		      \n\t"	// 4 * y
+	"vmovups  %%xmm7,-16(%4,%0,4)		      \n\t"	// 4 * y

 	"subq	        $16, %1			      \n\t"		
 	"jnz		1b		      \n\t"
@ -184,15 +184,15 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO

 	:
          "+r" (i),	// 0	
-	  "+r" (n)  	// 1
+	  "+r" (n),  	// 1
+          "+r" (lda4)   // 2
        : 
-          "r" (x),      // 2
-          "r" (y),      // 3
-          "r" (ap[0]),  // 4
-          "r" (ap[1]),  // 5
-          "r" (ap[2]),  // 6
-          "r" (ap[3]),  // 7
-          "r" (lda4),   // 8
+          "r" (x),      // 3
+          "r" (y),      // 4
+          "r" (ap[0]),  // 5
+          "r" (ap[1]),  // 6
+          "r" (ap[2]),  // 7
+          "r" (ap[3]),  // 8
          "r" (alpha)   // 9
 	: "cc", 
 	  "%xmm0", "%xmm1", 
--- a/kernel/x86_64/sgemv_n_microk_haswell-4.c
+++ b/kernel/x86_64/sgemv_n_microk_haswell-4.c
@ -26,7 +26,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *****************************************************************************/


-
 #define HAVE_KERNEL_4x8 1
 static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLONG lda4, FLOAT *alpha) __attribute__ ((noinline));

@ -38,41 +37,41 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	__asm__  __volatile__
 	(
 	"vzeroupper			 \n\t"
-	"vbroadcastss    (%2), %%ymm12	 \n\t"	// x0 
-	"vbroadcastss   4(%2), %%ymm13	 \n\t"	// x1 
-	"vbroadcastss   8(%2), %%ymm14	 \n\t"	// x2 
-	"vbroadcastss  12(%2), %%ymm15	 \n\t"	// x3 
-	"vbroadcastss  16(%2), %%ymm0 	 \n\t"	// x4 
-	"vbroadcastss  20(%2), %%ymm1 	 \n\t"	// x5 
-	"vbroadcastss  24(%2), %%ymm2 	 \n\t"	// x6 
-	"vbroadcastss  28(%2), %%ymm3 	 \n\t"	// x7 
+	"vbroadcastss    (%3), %%ymm12	 \n\t"	// x0 
+	"vbroadcastss   4(%3), %%ymm13	 \n\t"	// x1 
+	"vbroadcastss   8(%3), %%ymm14	 \n\t"	// x2 
+	"vbroadcastss  12(%3), %%ymm15	 \n\t"	// x3 
+	"vbroadcastss  16(%3), %%ymm0 	 \n\t"	// x4 
+	"vbroadcastss  20(%3), %%ymm1 	 \n\t"	// x5 
+	"vbroadcastss  24(%3), %%ymm2 	 \n\t"	// x6 
+	"vbroadcastss  28(%3), %%ymm3 	 \n\t"	// x7 

 	"vbroadcastss    (%9), %%ymm6 	 \n\t"	// alpha 

        "testq          $0x04, %1                      \n\t"
        "jz             2f                    \n\t"

-	"vmovups	(%3,%0,4), %%xmm7	       \n\t"	// 4 * y
+	"vmovups	(%4,%0,4), %%xmm7	       \n\t"	// 4 * y
 	"vxorps		%%xmm4 , %%xmm4, %%xmm4        \n\t"
 	"vxorps		%%xmm5 , %%xmm5, %%xmm5        \n\t"

-	"vfmadd231ps   (%4,%0,4), %%xmm12, %%xmm4      \n\t" 
-	"vfmadd231ps   (%5,%0,4), %%xmm13, %%xmm5      \n\t" 
-	"vfmadd231ps   (%6,%0,4), %%xmm14, %%xmm4      \n\t" 
-	"vfmadd231ps   (%7,%0,4), %%xmm15, %%xmm5      \n\t" 
+	"vfmadd231ps   (%5,%0,4), %%xmm12, %%xmm4      \n\t" 
+	"vfmadd231ps   (%6,%0,4), %%xmm13, %%xmm5      \n\t" 
+	"vfmadd231ps   (%7,%0,4), %%xmm14, %%xmm4      \n\t" 
+	"vfmadd231ps   (%8,%0,4), %%xmm15, %%xmm5      \n\t" 

-	"vfmadd231ps   (%4,%8,4), %%xmm0 , %%xmm4      \n\t" 
-	"vfmadd231ps   (%5,%8,4), %%xmm1 , %%xmm5      \n\t" 
-	"vfmadd231ps   (%6,%8,4), %%xmm2 , %%xmm4      \n\t" 
-	"vfmadd231ps   (%7,%8,4), %%xmm3 , %%xmm5      \n\t" 
+	"vfmadd231ps   (%5,%2,4), %%xmm0 , %%xmm4      \n\t" 
+	"vfmadd231ps   (%6,%2,4), %%xmm1 , %%xmm5      \n\t" 
+	"vfmadd231ps   (%7,%2,4), %%xmm2 , %%xmm4      \n\t" 
+	"vfmadd231ps   (%8,%2,4), %%xmm3 , %%xmm5      \n\t" 

 	"vaddps		%%xmm4 , %%xmm5 , %%xmm5       \n\t"
 	"vmulps		%%xmm6 , %%xmm5 , %%xmm5       \n\t"
 	"vaddps		%%xmm7 , %%xmm5 , %%xmm5       \n\t"

-	"vmovups  %%xmm5,   (%3,%0,4)		       \n\t"	// 4 * y
+	"vmovups  %%xmm5,   (%4,%0,4)		       \n\t"	// 4 * y

-        "addq		$4 , %8	  	 	       \n\t"
+        "addq		$4 , %2	  	 	       \n\t"
        "addq		$4 , %0	  	 	       \n\t"
 	"subq	        $4 , %1			       \n\t"		

@ -81,28 +80,28 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
        "testq          $0x08, %1                      \n\t"
        "jz             3f                    \n\t"

-	"vmovups	(%3,%0,4), %%ymm7	       \n\t"	// 8 * y
+	"vmovups	(%4,%0,4), %%ymm7	       \n\t"	// 8 * y
 	"vxorps		%%ymm4 , %%ymm4, %%ymm4        \n\t"
 	"vxorps		%%ymm5 , %%ymm5, %%ymm5        \n\t"

-	"vfmadd231ps   (%4,%0,4), %%ymm12, %%ymm4      \n\t" 
-	"vfmadd231ps   (%5,%0,4), %%ymm13, %%ymm5      \n\t" 
-	"vfmadd231ps   (%6,%0,4), %%ymm14, %%ymm4      \n\t" 
-	"vfmadd231ps   (%7,%0,4), %%ymm15, %%ymm5      \n\t" 
+	"vfmadd231ps   (%5,%0,4), %%ymm12, %%ymm4      \n\t" 
+	"vfmadd231ps   (%6,%0,4), %%ymm13, %%ymm5      \n\t" 
+	"vfmadd231ps   (%7,%0,4), %%ymm14, %%ymm4      \n\t" 
+	"vfmadd231ps   (%8,%0,4), %%ymm15, %%ymm5      \n\t" 

-	"vfmadd231ps   (%4,%8,4), %%ymm0 , %%ymm4      \n\t" 
-	"vfmadd231ps   (%5,%8,4), %%ymm1 , %%ymm5      \n\t" 
-	"vfmadd231ps   (%6,%8,4), %%ymm2 , %%ymm4      \n\t" 
-	"vfmadd231ps   (%7,%8,4), %%ymm3 , %%ymm5      \n\t" 
+	"vfmadd231ps   (%5,%2,4), %%ymm0 , %%ymm4      \n\t" 
+	"vfmadd231ps   (%6,%2,4), %%ymm1 , %%ymm5      \n\t" 
+	"vfmadd231ps   (%7,%2,4), %%ymm2 , %%ymm4      \n\t" 
+	"vfmadd231ps   (%8,%2,4), %%ymm3 , %%ymm5      \n\t" 

 	"vaddps		%%ymm4 , %%ymm5 , %%ymm5       \n\t"
 	"vmulps		%%ymm6 , %%ymm5 , %%ymm5       \n\t"
 	"vaddps		%%ymm7 , %%ymm5 , %%ymm5       \n\t"


-	"vmovups  %%ymm5,   (%3,%0,4)		       \n\t"	// 8 * y
+	"vmovups  %%ymm5,   (%4,%0,4)		       \n\t"	// 8 * y

-        "addq		$8 , %8	  	 	       \n\t"
+        "addq		$8 , %2	  	 	       \n\t"
        "addq		$8 , %0	  	 	       \n\t"
 	"subq	        $8 , %1			       \n\t"		

@ -117,35 +116,35 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO

 	"vxorps		%%ymm4 , %%ymm4, %%ymm4        \n\t"
 	"vxorps		%%ymm5 , %%ymm5, %%ymm5        \n\t"
-	"vmovups	(%3,%0,4), %%ymm8	       \n\t"	// 8 * y
-	"vmovups      32(%3,%0,4), %%ymm9	       \n\t"	// 8 * y
+	"vmovups	(%4,%0,4), %%ymm8	       \n\t"	// 8 * y
+	"vmovups      32(%4,%0,4), %%ymm9	       \n\t"	// 8 * y

-	"vfmadd231ps   (%4,%0,4), %%ymm12, %%ymm4      \n\t" 
-	"vfmadd231ps 32(%4,%0,4), %%ymm12, %%ymm5      \n\t" 
-	"vfmadd231ps   (%5,%0,4), %%ymm13, %%ymm4      \n\t" 
-	"vfmadd231ps 32(%5,%0,4), %%ymm13, %%ymm5      \n\t" 
-	"vfmadd231ps   (%6,%0,4), %%ymm14, %%ymm4      \n\t" 
-	"vfmadd231ps 32(%6,%0,4), %%ymm14, %%ymm5      \n\t" 
-	"vfmadd231ps   (%7,%0,4), %%ymm15, %%ymm4      \n\t" 
-	"vfmadd231ps 32(%7,%0,4), %%ymm15, %%ymm5      \n\t" 
+	"vfmadd231ps   (%5,%0,4), %%ymm12, %%ymm4      \n\t" 
+	"vfmadd231ps 32(%5,%0,4), %%ymm12, %%ymm5      \n\t" 
+	"vfmadd231ps   (%6,%0,4), %%ymm13, %%ymm4      \n\t" 
+	"vfmadd231ps 32(%6,%0,4), %%ymm13, %%ymm5      \n\t" 
+	"vfmadd231ps   (%7,%0,4), %%ymm14, %%ymm4      \n\t" 
+	"vfmadd231ps 32(%7,%0,4), %%ymm14, %%ymm5      \n\t" 
+	"vfmadd231ps   (%8,%0,4), %%ymm15, %%ymm4      \n\t" 
+	"vfmadd231ps 32(%8,%0,4), %%ymm15, %%ymm5      \n\t" 

-	"vfmadd231ps   (%4,%8,4), %%ymm0 , %%ymm4      \n\t" 
+	"vfmadd231ps   (%5,%2,4), %%ymm0 , %%ymm4      \n\t" 
        "addq		$16, %0	  	 	       \n\t"
-	"vfmadd231ps 32(%4,%8,4), %%ymm0 , %%ymm5      \n\t" 
-	"vfmadd231ps   (%5,%8,4), %%ymm1 , %%ymm4      \n\t" 
-	"vfmadd231ps 32(%5,%8,4), %%ymm1 , %%ymm5      \n\t" 
-	"vfmadd231ps   (%6,%8,4), %%ymm2 , %%ymm4      \n\t" 
-	"vfmadd231ps 32(%6,%8,4), %%ymm2 , %%ymm5      \n\t" 
-	"vfmadd231ps   (%7,%8,4), %%ymm3 , %%ymm4      \n\t" 
-	"vfmadd231ps 32(%7,%8,4), %%ymm3 , %%ymm5      \n\t" 
+	"vfmadd231ps 32(%5,%2,4), %%ymm0 , %%ymm5      \n\t" 
+	"vfmadd231ps   (%6,%2,4), %%ymm1 , %%ymm4      \n\t" 
+	"vfmadd231ps 32(%6,%2,4), %%ymm1 , %%ymm5      \n\t" 
+	"vfmadd231ps   (%7,%2,4), %%ymm2 , %%ymm4      \n\t" 
+	"vfmadd231ps 32(%7,%2,4), %%ymm2 , %%ymm5      \n\t" 
+	"vfmadd231ps   (%8,%2,4), %%ymm3 , %%ymm4      \n\t" 
+	"vfmadd231ps 32(%8,%2,4), %%ymm3 , %%ymm5      \n\t" 

 	"vfmadd231ps     %%ymm6 , %%ymm4 , %%ymm8      \n\t"
 	"vfmadd231ps     %%ymm6 , %%ymm5 , %%ymm9      \n\t"

-        "addq		$16, %8	  	 	      \n\t"
-	"vmovups  %%ymm8,-64(%3,%0,4)		      \n\t"	// 8 * y
+        "addq		$16, %2	  	 	      \n\t"
+	"vmovups  %%ymm8,-64(%4,%0,4)		      \n\t"	// 8 * y
 	"subq	        $16, %1			      \n\t"		
-	"vmovups  %%ymm9,-32(%3,%0,4)		      \n\t"	// 8 * y
+	"vmovups  %%ymm9,-32(%4,%0,4)		      \n\t"	// 8 * y

 	"jnz		1b		      \n\t"

@ -154,15 +153,15 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO

 	:
          "+r" (i),	// 0	
-	  "+r" (n)  	// 1
+	  "+r" (n),  	// 1
+          "+r" (lda4)   // 2
        :  
-	  "r" (x),      // 2
-          "r" (y),      // 3
-          "r" (ap[0]),  // 4
-          "r" (ap[1]),  // 5
-          "r" (ap[2]),  // 6
-          "r" (ap[3]),  // 7
-          "r" (lda4),   // 8
+	  "r" (x),      // 3
+          "r" (y),      // 4
+          "r" (ap[0]),  // 5
+          "r" (ap[1]),  // 6
+          "r" (ap[2]),  // 7
+          "r" (ap[3]),  // 8
          "r" (alpha)   // 9
 	: "cc", 
 	  "%xmm0", "%xmm1", 
@ -177,7 +176,6 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 } 


-
 #define HAVE_KERNEL_4x4 1
 static void sgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha) __attribute__ ((noinline));

@ -196,6 +194,7 @@ static void sgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT

 	"vbroadcastss    (%8), %%ymm6 	 \n\t"	// alpha 

+
        "testq          $0x04, %1                      \n\t"
        "jz             2f                    \n\t"

--- a/kernel/x86_64/sgemv_n_microk_nehalem-4.c
+++ b/kernel/x86_64/sgemv_n_microk_nehalem-4.c
@ -37,19 +37,19 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO

 	__asm__  __volatile__
 	(
-	"movss    (%2), %%xmm12	 \n\t"	// x0 
-	"movss   4(%2), %%xmm13	 \n\t"	// x1 
-	"movss   8(%2), %%xmm14	 \n\t"	// x2 
-	"movss  12(%2), %%xmm15	 \n\t"	// x3 
+	"movss    (%3), %%xmm12	 \n\t"	// x0 
+	"movss   4(%3), %%xmm13	 \n\t"	// x1 
+	"movss   8(%3), %%xmm14	 \n\t"	// x2 
+	"movss  12(%3), %%xmm15	 \n\t"	// x3 
 	"shufps $0,  %%xmm12, %%xmm12\n\t"	
 	"shufps $0,  %%xmm13, %%xmm13\n\t"	
 	"shufps $0,  %%xmm14, %%xmm14\n\t"	
 	"shufps $0,  %%xmm15, %%xmm15\n\t"	

-	"movss  16(%2), %%xmm0	 \n\t"	// x4 
-	"movss  20(%2), %%xmm1	 \n\t"	// x5 
-	"movss  24(%2), %%xmm2	 \n\t"	// x6 
-	"movss  28(%2), %%xmm3	 \n\t"	// x7 
+	"movss  16(%3), %%xmm0	 \n\t"	// x4 
+	"movss  20(%3), %%xmm1	 \n\t"	// x5 
+	"movss  24(%3), %%xmm2	 \n\t"	// x6 
+	"movss  28(%3), %%xmm3	 \n\t"	// x7 
 	"shufps $0,  %%xmm0 , %%xmm0 \n\t"	
 	"shufps $0,  %%xmm1 , %%xmm1 \n\t"	
 	"shufps $0,  %%xmm2 , %%xmm2 \n\t"	
@ -63,13 +63,13 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"1:				 \n\t"
 	"xorps           %%xmm4 , %%xmm4	 \n\t"
 	"xorps           %%xmm5 , %%xmm5	 \n\t"
-	"movups             (%3,%0,4), %%xmm7          \n\t" // 4 * y
+	"movups             (%4,%0,4), %%xmm7          \n\t" // 4 * y

 	".p2align 1				       \n\t"
-	"movups             (%4,%0,4), %%xmm8          \n\t" 
-	"movups             (%5,%0,4), %%xmm9          \n\t" 
-	"movups             (%6,%0,4), %%xmm10         \n\t" 
-	"movups             (%7,%0,4), %%xmm11         \n\t" 
+	"movups             (%5,%0,4), %%xmm8          \n\t" 
+	"movups             (%6,%0,4), %%xmm9          \n\t" 
+	"movups             (%7,%0,4), %%xmm10         \n\t" 
+	"movups             (%8,%0,4), %%xmm11         \n\t" 
 	".p2align 1				       \n\t"
 	"mulps		%%xmm12, %%xmm8		       \n\t"
 	"mulps		%%xmm13, %%xmm9		       \n\t"
@ -80,10 +80,10 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"addps		%%xmm10, %%xmm4	               \n\t"
 	"addps		%%xmm11, %%xmm5 	       \n\t"

-	"movups             (%4,%8,4), %%xmm8          \n\t" 
-	"movups             (%5,%8,4), %%xmm9          \n\t" 
-	"movups             (%6,%8,4), %%xmm10         \n\t" 
-	"movups             (%7,%8,4), %%xmm11         \n\t" 
+	"movups             (%5,%2,4), %%xmm8          \n\t" 
+	"movups             (%6,%2,4), %%xmm9          \n\t" 
+	"movups             (%7,%2,4), %%xmm10         \n\t" 
+	"movups             (%8,%2,4), %%xmm11         \n\t" 
 	".p2align 1				       \n\t"
 	"mulps		%%xmm0 , %%xmm8		       \n\t"
 	"mulps		%%xmm1 , %%xmm9		       \n\t"
@ -94,28 +94,28 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"addps		%%xmm10, %%xmm4	       	       \n\t"
 	"addps		%%xmm11, %%xmm5 	       \n\t"

-        "addq		$4 , %8	  	 	       \n\t"
+        "addq		$4 , %2	  	 	       \n\t"
 	"addps		%%xmm5 , %%xmm4 	       \n\t"
        "addq		$4 , %0	  	 	       \n\t"
 	"mulps		%%xmm6 , %%xmm4		       \n\t" 
 	"subq	        $4 , %1			       \n\t"		
 	"addps		%%xmm4 , %%xmm7 	       \n\t"

-	"movups  %%xmm7 , -16(%3,%0,4)		       \n\t"	// 4 * y
+	"movups  %%xmm7 , -16(%4,%0,4)		       \n\t"	// 4 * y

 	"jnz		1b		       \n\t"

 	:
          "+r" (i),	// 0	
-	  "+r" (n)  	// 1
+	  "+r" (n), 	// 1
+          "+r" (lda4)   // 2
        : 
-          "r" (x),      // 2
-          "r" (y),      // 3
-          "r" (ap[0]),  // 4
-          "r" (ap[1]),  // 5
-          "r" (ap[2]),  // 6
-          "r" (ap[3]),  // 7
-          "r" (lda4),   // 8
+          "r" (x),      // 3
+          "r" (y),      // 4
+          "r" (ap[0]),  // 5
+          "r" (ap[1]),  // 6
+          "r" (ap[2]),  // 7
+          "r" (ap[3]),  // 8
          "r" (alpha)   // 9
 	: "cc", 
 	  "%xmm0", "%xmm1", 
--- a/kernel/x86_64/sgemv_n_microk_sandy-4.c
+++ b/kernel/x86_64/sgemv_n_microk_sandy-4.c
@ -39,14 +39,14 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	__asm__  __volatile__
 	(
 	"vzeroupper			 \n\t"
-	"vbroadcastss    (%2), %%ymm12	 \n\t"	// x0 
-	"vbroadcastss   4(%2), %%ymm13	 \n\t"	// x1 
-	"vbroadcastss   8(%2), %%ymm14	 \n\t"	// x2 
-	"vbroadcastss  12(%2), %%ymm15	 \n\t"	// x3 
-	"vbroadcastss  16(%2), %%ymm0 	 \n\t"	// x4 
-	"vbroadcastss  20(%2), %%ymm1 	 \n\t"	// x5 
-	"vbroadcastss  24(%2), %%ymm2 	 \n\t"	// x6 
-	"vbroadcastss  28(%2), %%ymm3 	 \n\t"	// x7 
+	"vbroadcastss    (%3), %%ymm12	 \n\t"	// x0 
+	"vbroadcastss   4(%3), %%ymm13	 \n\t"	// x1 
+	"vbroadcastss   8(%3), %%ymm14	 \n\t"	// x2 
+	"vbroadcastss  12(%3), %%ymm15	 \n\t"	// x3 
+	"vbroadcastss  16(%3), %%ymm0 	 \n\t"	// x4 
+	"vbroadcastss  20(%3), %%ymm1 	 \n\t"	// x5 
+	"vbroadcastss  24(%3), %%ymm2 	 \n\t"	// x6 
+	"vbroadcastss  28(%3), %%ymm3 	 \n\t"	// x7 

 	"vbroadcastss    (%9), %%ymm6 	 \n\t"	// alpha 

@ -55,21 +55,21 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO

 	"vxorps	  %%xmm4 , %%xmm4 , %%xmm4        \n\t"
 	"vxorps	  %%xmm5 , %%xmm5 , %%xmm5        \n\t"
-	"vmovups	(%3,%0,4), %%xmm7	  \n\t"	// 4 * y
+	"vmovups	(%4,%0,4), %%xmm7	  \n\t"	// 4 * y

-	"vmulps   (%4,%0,4), %%xmm12, %%xmm8      \n\t" 
-	"vmulps   (%5,%0,4), %%xmm13, %%xmm10     \n\t" 
-	"vmulps   (%6,%0,4), %%xmm14, %%xmm9      \n\t" 
-	"vmulps   (%7,%0,4), %%xmm15, %%xmm11     \n\t" 
+	"vmulps   (%5,%0,4), %%xmm12, %%xmm8      \n\t" 
+	"vmulps   (%6,%0,4), %%xmm13, %%xmm10     \n\t" 
+	"vmulps   (%7,%0,4), %%xmm14, %%xmm9      \n\t" 
+	"vmulps   (%8,%0,4), %%xmm15, %%xmm11     \n\t" 
 	"vaddps	  %%xmm4, %%xmm8 , %%xmm4	  \n\t"
 	"vaddps	  %%xmm5, %%xmm10, %%xmm5	  \n\t"
 	"vaddps	  %%xmm4, %%xmm9 , %%xmm4	  \n\t"
 	"vaddps	  %%xmm5, %%xmm11, %%xmm5	  \n\t"

-	"vmulps   (%4,%8,4), %%xmm0 , %%xmm8      \n\t" 
-	"vmulps   (%5,%8,4), %%xmm1 , %%xmm10     \n\t" 
-	"vmulps   (%6,%8,4), %%xmm2 , %%xmm9      \n\t" 
-	"vmulps   (%7,%8,4), %%xmm3 , %%xmm11     \n\t" 
+	"vmulps   (%5,%2,4), %%xmm0 , %%xmm8      \n\t" 
+	"vmulps   (%6,%2,4), %%xmm1 , %%xmm10     \n\t" 
+	"vmulps   (%7,%2,4), %%xmm2 , %%xmm9      \n\t" 
+	"vmulps   (%8,%2,4), %%xmm3 , %%xmm11     \n\t" 
 	"vaddps	  %%xmm4, %%xmm8 , %%xmm4	  \n\t"
 	"vaddps	  %%xmm5, %%xmm10, %%xmm5	  \n\t"
 	"vaddps	  %%xmm4, %%xmm9 , %%xmm4	  \n\t"
@ -79,9 +79,9 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"vmulps	  %%xmm6, %%xmm4 , %%xmm5	  \n\t"
 	"vaddps	  %%xmm5, %%xmm7 , %%xmm5	  \n\t"

-	"vmovups  %%xmm5,   (%3,%0,4)		  \n\t"	// 4 * y
+	"vmovups  %%xmm5,   (%4,%0,4)		  \n\t"	// 4 * y

-        "addq		$4, %8	  	 	  \n\t"
+        "addq		$4, %2	  	 	  \n\t"
        "addq		$4, %0	  	 	  \n\t"
 	"subq	        $4, %1			  \n\t"		

@ -92,21 +92,21 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO

 	"vxorps	  %%ymm4 , %%ymm4 , %%ymm4        \n\t"
 	"vxorps	  %%ymm5 , %%ymm5 , %%ymm5        \n\t"
-	"vmovups	(%3,%0,4), %%ymm7	  \n\t"	// 8 * y
+	"vmovups	(%4,%0,4), %%ymm7	  \n\t"	// 8 * y

-	"vmulps   (%4,%0,4), %%ymm12, %%ymm8      \n\t" 
-	"vmulps   (%5,%0,4), %%ymm13, %%ymm10     \n\t" 
-	"vmulps   (%6,%0,4), %%ymm14, %%ymm9      \n\t" 
-	"vmulps   (%7,%0,4), %%ymm15, %%ymm11     \n\t" 
+	"vmulps   (%5,%0,4), %%ymm12, %%ymm8      \n\t" 
+	"vmulps   (%6,%0,4), %%ymm13, %%ymm10     \n\t" 
+	"vmulps   (%7,%0,4), %%ymm14, %%ymm9      \n\t" 
+	"vmulps   (%8,%0,4), %%ymm15, %%ymm11     \n\t" 
 	"vaddps	  %%ymm4, %%ymm8 , %%ymm4	  \n\t"
 	"vaddps	  %%ymm5, %%ymm10, %%ymm5	  \n\t"
 	"vaddps	  %%ymm4, %%ymm9 , %%ymm4	  \n\t"
 	"vaddps	  %%ymm5, %%ymm11, %%ymm5	  \n\t"

-	"vmulps   (%4,%8,4), %%ymm0 , %%ymm8      \n\t" 
-	"vmulps   (%5,%8,4), %%ymm1 , %%ymm10     \n\t" 
-	"vmulps   (%6,%8,4), %%ymm2 , %%ymm9      \n\t" 
-	"vmulps   (%7,%8,4), %%ymm3 , %%ymm11     \n\t" 
+	"vmulps   (%5,%2,4), %%ymm0 , %%ymm8      \n\t" 
+	"vmulps   (%6,%2,4), %%ymm1 , %%ymm10     \n\t" 
+	"vmulps   (%7,%2,4), %%ymm2 , %%ymm9      \n\t" 
+	"vmulps   (%8,%2,4), %%ymm3 , %%ymm11     \n\t" 
 	"vaddps	  %%ymm4, %%ymm8 , %%ymm4	  \n\t"
 	"vaddps	  %%ymm5, %%ymm10, %%ymm5	  \n\t"
 	"vaddps	  %%ymm4, %%ymm9 , %%ymm4	  \n\t"
@ -116,9 +116,9 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"vmulps	  %%ymm6, %%ymm4 , %%ymm5	  \n\t"
 	"vaddps	  %%ymm5, %%ymm7 , %%ymm5	  \n\t"

-	"vmovups  %%ymm5,   (%3,%0,4)		  \n\t"	// 8 * y
+	"vmovups  %%ymm5,   (%4,%0,4)		  \n\t"	// 8 * y

-        "addq		$8, %8	  	 	  \n\t"
+        "addq		$8, %2	  	 	  \n\t"
        "addq		$8, %0	  	 	  \n\t"
 	"subq	        $8, %1			  \n\t"		

@ -134,45 +134,45 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"vxorps	  %%ymm4 , %%ymm4 , %%ymm4        \n\t"
 	"vxorps	  %%ymm5 , %%ymm5 , %%ymm5        \n\t"

-	"prefetcht0	 192(%4,%0,4)		  \n\t"
-	"vmulps   (%4,%0,4), %%ymm12, %%ymm8      \n\t" 
-	"vmulps 32(%4,%0,4), %%ymm12, %%ymm9      \n\t" 
 	"prefetcht0	 192(%5,%0,4)		  \n\t"
-	"vmulps   (%5,%0,4), %%ymm13, %%ymm10     \n\t" 
-	"vmulps 32(%5,%0,4), %%ymm13, %%ymm11     \n\t" 
-	"vaddps	  %%ymm4, %%ymm8 , %%ymm4	  \n\t"
-	"vaddps	  %%ymm5, %%ymm9 , %%ymm5	  \n\t"
-	"vaddps	  %%ymm4, %%ymm10, %%ymm4	  \n\t"
-	"vaddps	  %%ymm5, %%ymm11, %%ymm5	  \n\t"
-
+	"vmulps   (%5,%0,4), %%ymm12, %%ymm8      \n\t" 
+	"vmulps 32(%5,%0,4), %%ymm12, %%ymm9      \n\t" 
 	"prefetcht0	 192(%6,%0,4)		  \n\t"
-	"vmulps   (%6,%0,4), %%ymm14, %%ymm8      \n\t" 
-	"vmulps 32(%6,%0,4), %%ymm14, %%ymm9      \n\t" 
+	"vmulps   (%6,%0,4), %%ymm13, %%ymm10     \n\t" 
+	"vmulps 32(%6,%0,4), %%ymm13, %%ymm11     \n\t" 
+	"vaddps	  %%ymm4, %%ymm8 , %%ymm4	  \n\t"
+	"vaddps	  %%ymm5, %%ymm9 , %%ymm5	  \n\t"
+	"vaddps	  %%ymm4, %%ymm10, %%ymm4	  \n\t"
+	"vaddps	  %%ymm5, %%ymm11, %%ymm5	  \n\t"
+
 	"prefetcht0	 192(%7,%0,4)		  \n\t"
-	"vmulps   (%7,%0,4), %%ymm15, %%ymm10     \n\t" 
-	"vmulps 32(%7,%0,4), %%ymm15, %%ymm11     \n\t" 
+	"vmulps   (%7,%0,4), %%ymm14, %%ymm8      \n\t" 
+	"vmulps 32(%7,%0,4), %%ymm14, %%ymm9      \n\t" 
+	"prefetcht0	 192(%8,%0,4)		  \n\t"
+	"vmulps   (%8,%0,4), %%ymm15, %%ymm10     \n\t" 
+	"vmulps 32(%8,%0,4), %%ymm15, %%ymm11     \n\t" 
 	"vaddps	  %%ymm4, %%ymm8 , %%ymm4	  \n\t"
 	"vaddps	  %%ymm5, %%ymm9 , %%ymm5	  \n\t"
 	"vaddps	  %%ymm4, %%ymm10, %%ymm4	  \n\t"
 	"vaddps	  %%ymm5, %%ymm11, %%ymm5	  \n\t"

-	"prefetcht0	 192(%4,%8,4)		  \n\t"
-	"vmulps   (%4,%8,4), %%ymm0 , %%ymm8      \n\t" 
-	"vmulps 32(%4,%8,4), %%ymm0 , %%ymm9      \n\t" 
-	"prefetcht0	 192(%5,%8,4)		  \n\t"
-	"vmulps   (%5,%8,4), %%ymm1 , %%ymm10     \n\t" 
-	"vmulps 32(%5,%8,4), %%ymm1 , %%ymm11     \n\t" 
+	"prefetcht0	 192(%5,%2,4)		  \n\t"
+	"vmulps   (%5,%2,4), %%ymm0 , %%ymm8      \n\t" 
+	"vmulps 32(%5,%2,4), %%ymm0 , %%ymm9      \n\t" 
+	"prefetcht0	 192(%6,%2,4)		  \n\t"
+	"vmulps   (%6,%2,4), %%ymm1 , %%ymm10     \n\t" 
+	"vmulps 32(%6,%2,4), %%ymm1 , %%ymm11     \n\t" 
 	"vaddps	  %%ymm4, %%ymm8 , %%ymm4	  \n\t"
 	"vaddps	  %%ymm5, %%ymm9 , %%ymm5	  \n\t"
 	"vaddps	  %%ymm4, %%ymm10, %%ymm4	  \n\t"
 	"vaddps	  %%ymm5, %%ymm11, %%ymm5	  \n\t"

-	"prefetcht0	 192(%6,%8,4)		  \n\t"
-	"vmulps   (%6,%8,4), %%ymm2 , %%ymm8      \n\t" 
-	"vmulps 32(%6,%8,4), %%ymm2 , %%ymm9      \n\t" 
-	"prefetcht0	 192(%7,%8,4)		  \n\t"
-	"vmulps   (%7,%8,4), %%ymm3 , %%ymm10     \n\t" 
-	"vmulps 32(%7,%8,4), %%ymm3 , %%ymm11     \n\t" 
+	"prefetcht0	 192(%7,%2,4)		  \n\t"
+	"vmulps   (%7,%2,4), %%ymm2 , %%ymm8      \n\t" 
+	"vmulps 32(%7,%2,4), %%ymm2 , %%ymm9      \n\t" 
+	"prefetcht0	 192(%8,%2,4)		  \n\t"
+	"vmulps   (%8,%2,4), %%ymm3 , %%ymm10     \n\t" 
+	"vmulps 32(%8,%2,4), %%ymm3 , %%ymm11     \n\t" 
 	"vaddps	  %%ymm4, %%ymm8 , %%ymm4	  \n\t"
 	"vaddps	  %%ymm5, %%ymm9 , %%ymm5	  \n\t"
 	"vaddps	  %%ymm4, %%ymm10, %%ymm4	  \n\t"
@ -181,13 +181,13 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"vmulps	  %%ymm6, %%ymm4 , %%ymm4	  \n\t"
 	"vmulps	  %%ymm6, %%ymm5 , %%ymm5	  \n\t"

-	"vaddps    (%3,%0,4), %%ymm4 , %%ymm4	 \n\t"	// 8 * y
-	"vaddps  32(%3,%0,4), %%ymm5 , %%ymm5	 \n\t"	// 8 * y
+	"vaddps    (%4,%0,4), %%ymm4 , %%ymm4	 \n\t"	// 8 * y
+	"vaddps  32(%4,%0,4), %%ymm5 , %%ymm5	 \n\t"	// 8 * y

-	"vmovups  %%ymm4,   (%3,%0,4)		  \n\t"	// 8 * y
-	"vmovups  %%ymm5, 32(%3,%0,4)		  \n\t"	// 8 * y
+	"vmovups  %%ymm4,   (%4,%0,4)		  \n\t"	// 8 * y
+	"vmovups  %%ymm5, 32(%4,%0,4)		  \n\t"	// 8 * y

-        "addq		$16, %8	  	 	  \n\t"
+        "addq		$16, %2	  	 	  \n\t"
        "addq		$16, %0	  	 	  \n\t"
 	"subq	        $16, %1			  \n\t"		
 	"jnz		1b		  \n\t"
@ -197,15 +197,15 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO

 	:
          "+r" (i),	// 0	
-	  "+r" (n)  	// 1
+	  "+r" (n),  	// 1
+          "+r" (lda4)   // 2
        : 
-          "r" (x),      // 2
-          "r" (y),      // 3
-          "r" (ap[0]),  // 4
-          "r" (ap[1]),  // 5
-          "r" (ap[2]),  // 6
-          "r" (ap[3]),  // 7
-          "r" (lda4),   // 8
+          "r" (x),      // 3
+          "r" (y),      // 4
+          "r" (ap[0]),  // 5
+          "r" (ap[1]),  // 6
+          "r" (ap[2]),  // 7
+          "r" (ap[3]),  // 8
          "r" (alpha)   // 9
 	: "cc", 
 	  "%xmm0", "%xmm1", 
--- a/kernel/x86_64/sgemv_t_4.c
+++ b/kernel/x86_64/sgemv_t_4.c
@ -139,9 +139,9 @@ static void sgemv_kernel_4x2(BLASLONG n, FLOAT *ap0, FLOAT *ap1, FLOAT *x, FLOAT
 	"movss	       %%xmm11,4(%2)	        \n\t"

        :
-   	:
-	"r" (i),	 // 0
-	"r" (n),	 // 1
+	"+r" (i),	 // 0
+	"+r" (n)	 // 1
+	:
        "r" (y),         // 2    
        "r" (ap0),       // 3
        "r" (ap1),       // 4
@ -208,9 +208,9 @@ static void sgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y)
 	"movss	       %%xmm10, (%2)	        \n\t"

        :
-   	:
-	"r" (i),	 // 0
-	"r" (n),	 // 1
+	"+r" (i),	 // 0
+	"+r" (n)	 // 1
+	:
        "r" (y),         // 2    
        "r" (ap),        // 3
        "r" (x)          // 4
@ -272,9 +272,9 @@ static void add_y(BLASLONG n, FLOAT da , FLOAT *src, FLOAT *dest, BLASLONG inc_d
        "jnz            1b              \n\t"

        :
-   	:
-	"r" (i),	  // 0
-	"r" (n),	  // 1
+	"+r" (i),	  // 0
+	"+r" (n)	  // 1
+	:
        "r" (&da),        // 2    
        "r" (src),        // 3
        "r" (dest)        // 4
--- a/Show More
+++ b/Show More