Define new VSE512 kernels

configure.ac

@@ -825,6 +825,14 @@ m4_define(elpa_m4_avx512_kernels, [
         complex_avx512_block2
 ])
 
+m4_define(elpa_m4_vse512_kernels, [
+        real_vse512_block2
+        real_vse512_block4
+        real_vse512_block6
+        complex_vse512_block1
+        complex_vse512_block2
+])
+
 m4_define(elpa_m4_bgp_kernels, [
         real_bgp
         complex_bgp
@@ -840,7 +848,7 @@ m4_define(elpa_m4_gpu_kernels, [
         complex_gpu
 ])
 
-m4_define(elpa_m4_kernel_types, [generic sparc64 neon_arch64 vsx sse sse_assembly avx avx2 avx512 bgp bgq gpu])
+m4_define(elpa_m4_kernel_types, [generic sparc64 neon_arch64 vsx sse sse_assembly avx avx2 avx512 vse512 bgp bgq gpu])
 
 m4_define(elpa_m4_all_kernels,
           m4_foreach_w([elpa_m4_type],
@@ -881,6 +889,7 @@ ELPA_SELECT_KERNELS([sse_assembly],[enable])
 ELPA_SELECT_KERNELS([avx],[enable])
 ELPA_SELECT_KERNELS([avx2],[enable])
 ELPA_SELECT_KERNELS([avx512],[enable])
+ELPA_SELECT_KERNELS([vse512],[disable])
 ELPA_SELECT_KERNELS([gpu],[disable])
 ELPA_SELECT_KERNELS([bgp],[disable])
 ELPA_SELECT_KERNELS([bgq],[disable])
@@ -890,7 +899,7 @@ m4_foreach_w([elpa_m4_kind],[real complex],[
 ])
 
 if test x"${enable_bgp}" = x"yes" -o x"$enable_bgq" = x"yes"; then
-        m4_foreach_w([elpa_m4_kernel], elpa_m4_sparc64_kernels elpa_m4_neon_arch64_kernels elpa_m4_vsx_kernels elpa_m4_sse_kernels elpa_m4_avx_kernels elpa_m4_avx2_kernels elpa_m4_avx512_kernels, [
+        m4_foreach_w([elpa_m4_kernel], elpa_m4_sparc64_kernels elpa_m4_neon_arch64_kernels elpa_m4_vsx_kernels elpa_m4_sse_kernels elpa_m4_avx_kernels elpa_m4_avx2_kernels elpa_m4_avx512_kernels elpa_m4_vse512_kernels, [
                 if x"$use_[]elpa_m4_kernel[]" = x"yes" ; then
                         echo "Disabling elpa_m4_kernel due to BGP/BGQ option"
                 fi
@@ -950,7 +959,7 @@ AC_DEFUN([ELPA_KERNEL_DEPENDS],[
                   ])
           fi
 ])
-m4_foreach_w([elpa_m4_arch],[sparc64 neon_arch64 vsx sse avx avx2 avx512],[
+m4_foreach_w([elpa_m4_arch],[sparc64 neon_arch64 vsx sse avx avx2 avx512 vse512],[
         ELPA_KERNEL_DEPENDS([real_]elpa_m4_arch[_block6], [real_]elpa_m4_arch[_block4 real_]elpa_m4_arch[_block2])
         ELPA_KERNEL_DEPENDS([real_]elpa_m4_arch[_block4], [real_]elpa_m4_arch[_block2])
         ELPA_KERNEL_DEPENDS([complex_]elpa_m4_arch[_block2], [complex_]elpa_m4_arch[_block1])
@@ -1008,7 +1017,7 @@ m4_foreach_w([elpa_m4_kind],[real complex],[
 m4_foreach_w([elpa_m4_kind],[real complex],[
         m4_foreach_w([elpa_m4_kernel],
                      m4_foreach_w([elpa_m4_cand_kernel],
-                                  elpa_m4_avx512_kernels elpa_m4_avx2_kernels elpa_m4_avx_kernels elpa_m4_sse_kernels elpa_m4_sse_assembly_kernels elpa_m4_sparc64_kernels elpa_m4_neon_arch64_kernels elpa_m4_vsx_kernels elpa_m4_generic_kernels elpa_m4_gpu_kernels,
+                                  elpa_m4_avx512_kernels elpa_m4_avx2_kernels elpa_m4_avx_kernels elpa_m4_sse_kernels elpa_m4_sse_assembly_kernels elpa_m4_vse512_kernels elpa_m4_sparc64_kernels elpa_m4_neon_arch64_kernels elpa_m4_vsx_kernels elpa_m4_generic_kernels elpa_m4_gpu_kernels,
                                   [m4_bmatch(elpa_m4_cand_kernel,elpa_m4_kind,elpa_m4_cand_kernel)] ),
                      [
                      if test -z "$default_[]elpa_m4_kind[]_kernel"; then

elpa/elpa_constants.h.in

@@ -60,8 +60,11 @@ enum ELPA_SOLVERS {
         X(ELPA_2STAGE_REAL_VSX_BLOCK2, 25, @ELPA_2STAGE_REAL_VSX_BLOCK2_COMPILED@, __VA_ARGS__) \
         X(ELPA_2STAGE_REAL_VSX_BLOCK4, 26, @ELPA_2STAGE_REAL_VSX_BLOCK4_COMPILED@, __VA_ARGS__) \
         X(ELPA_2STAGE_REAL_VSX_BLOCK6, 27, @ELPA_2STAGE_REAL_VSX_BLOCK6_COMPILED@, __VA_ARGS__) \
-        X(ELPA_2STAGE_REAL_GENERIC_SIMPLE_BLOCK4, 28, @ELPA_2STAGE_REAL_GENERIC_SIMPLE_BLOCK4_COMPILED@, __VA_ARGS__) \
-        X(ELPA_2STAGE_REAL_GENERIC_SIMPLE_BLOCK6, 29, @ELPA_2STAGE_REAL_GENERIC_SIMPLE_BLOCK6_COMPILED@, __VA_ARGS__)
+        X(ELPA_2STAGE_REAL_VSE512_BLOCK2, 28, @ELPA_2STAGE_REAL_VSE512_BLOCK2_COMPILED@, __VA_ARGS__) \
+        X(ELPA_2STAGE_REAL_VSE512_BLOCK4, 29, @ELPA_2STAGE_REAL_VSE512_BLOCK4_COMPILED@, __VA_ARGS__) \
+        X(ELPA_2STAGE_REAL_VSE512_BLOCK6, 30, @ELPA_2STAGE_REAL_VSE512_BLOCK6_COMPILED@, __VA_ARGS__) \
+        X(ELPA_2STAGE_REAL_GENERIC_SIMPLE_BLOCK4, 31, @ELPA_2STAGE_REAL_GENERIC_SIMPLE_BLOCK4_COMPILED@, __VA_ARGS__) \
+        X(ELPA_2STAGE_REAL_GENERIC_SIMPLE_BLOCK6, 32, @ELPA_2STAGE_REAL_GENERIC_SIMPLE_BLOCK6_COMPILED@, __VA_ARGS__)
 
 #define ELPA_FOR_ALL_2STAGE_REAL_KERNELS_AND_DEFAULT(X) \
         ELPA_FOR_ALL_2STAGE_REAL_KERNELS(X) \

elpa/elpa_simd_constants.h

@@ -9,5 +9,6 @@
 #define VSX_INSTR 9
 #define ARCH64_INSTR 10
 #define SPARC_INSTR 11
+#define VSE512_INSTR 12
 
-#define NUMBER_OF_INSTR 12
+#define NUMBER_OF_INSTR 13