Merge pull request #50 from xianyi/develop

rebase

.travis.yml

@@ -16,7 +16,6 @@ matrix:
       before_script: &common-before
         - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"
       script:
-        - set -e
         - make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
         - make -C test $COMMON_FLAGS $BTYPE
         - make -C ctest $COMMON_FLAGS $BTYPE
@@ -108,7 +107,6 @@ matrix:
         - sudo sh alpine-chroot-install -p 'build-base gfortran perl linux-headers'
       before_script: *common-before
       script:
-        - set -e
         # XXX: Disable some warnings for now to avoid exceeding Travis limit for log size.
         - alpine make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
               CFLAGS="-Wno-misleading-indentation -Wno-sign-conversion -Wno-incompatible-pointer-types"
@@ -151,7 +149,6 @@ matrix:
       before_script:
         - COMMON_ARGS="-DTARGET=NEHALEM -DNUM_THREADS=32"
       script:
-        - set -e
         - mkdir build
         - CONFIG=Release
         - cmake -Bbuild -H. $CMAKE_ARGS $COMMON_ARGS -DCMAKE_BUILD_TYPE=$CONFIG

CONTRIBUTORS.md

@@ -180,3 +180,7 @@ In chronological order:
   * [2019-12-23] optimize AVX2 CGEMM and ZGEMM
   * [2019-12-30] AVX2 CGEMM3M & ZGEMM3M kernels
   * [2020-01-07] optimize AVX2 SGEMM and STRMM
+
+* Rajalakshmi Srinivasaraghavan <https://github.com/RajalakshmiSR>
+  * [2020-04-15] Half-precision GEMM for bfloat16
+  

cmake/prebuild.cmake

@@ -492,7 +492,7 @@ else(NOT CMAKE_CROSSCOMPILING)
   if (NOT "${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
     try_compile(GETARCH_RESULT ${GETARCH_DIR}
       SOURCES ${GETARCH_SRC}
-    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} -I${GETARCH_DIR} -I"${PROJECT_SOURCE_DIR}" -I"${PROJECT_BINARY_DIR}"
+    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} -I"${GETARCH_DIR}" -I"${PROJECT_SOURCE_DIR}" -I"${PROJECT_BINARY_DIR}"
       OUTPUT_VARIABLE GETARCH_LOG
       COPY_FILE ${PROJECT_BINARY_DIR}/${GETARCH_BIN}
     )
@@ -520,7 +520,7 @@ execute_process(COMMAND "${PROJECT_BINARY_DIR}/${GETARCH_BIN}" 1 OUTPUT_VARIABLE
   if (NOT "${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
     try_compile(GETARCH2_RESULT ${GETARCH2_DIR}
       SOURCES ${PROJECT_SOURCE_DIR}/getarch_2nd.c
-    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} ${GETARCH2_FLAGS} -I${GETARCH2_DIR} -I"${PROJECT_SOURCE_DIR}" -I"${PROJECT_BINARY_DIR}"
+    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} ${GETARCH2_FLAGS} -I"${GETARCH2_DIR}" -I"${PROJECT_SOURCE_DIR}" -I"${PROJECT_BINARY_DIR}"
       OUTPUT_VARIABLE GETARCH2_LOG
       COPY_FILE ${PROJECT_BINARY_DIR}/${GETARCH2_BIN}
     )

param.h

@@ -2254,7 +2254,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SGEMM_DEFAULT_R 4096
 #define DGEMM_DEFAULT_R 4096
 #define CGEMM_DEFAULT_R 4096
-#define ZGEMM_DEFAULT_R 512
+#define ZGEMM_DEFAULT_R 4096
 
 #define SYMV_P	 8
 
@@ -2288,6 +2288,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CGEMM_DEFAULT_Q  1026
 #define ZGEMM_DEFAULT_Q 1026
 
+#define SGEMM_DEFAULT_R 4096
+#define DGEMM_DEFAULT_R 4096
+#define CGEMM_DEFAULT_R 4096
+#define ZGEMM_DEFAULT_R 4096
+
 #define SYMV_P	 8
 
 #endif