Single precision kernel for AVX512 real block6

.gitlab-ci.yml

@@ -2192,6 +2192,19 @@ intel-double-precision-mpi-noopenmp-ftimings-redirect-real-avx512_block6-complex
     - export LD_LIBRARY_PATH=$MKL_HOME/lib/intel64:$LD_LIBRARY_PATH
     -  make check TEST_FLAGS='1000 500 128'
 
+intel-single-precision-mpi-noopenmp-ftimings-redirect-real-avx512_block6-complex-avx512_block1-kernel-jobs:
+  tags:
+    - KNL
+  script:
+    - ./autogen.sh
+    - ./configure FC=mpiifort CC=mpiicc CFLAGS="-O3 -mtune=knl -axMIC-AVX512" FCFLAGS="-O3 -mtune=knl -axMIC-AVX512" SCALAPACK_FCFLAGS="-L$MKLROOT/lib/intel64 -lmkl_scalapack_lp64 -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lmkl_blacs_intelmpi_lp64 -lpthread -lm -I$MKLROOT/include/intel64/lp64" SCALAPACK_LDFLAGS="-L$MKLROOT/lib/intel64 -lmkl_scalapack_lp64 -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lmkl_blacs_intelmpi_lp64 -lpthread -lm -Wl,-rpath,$MKLROOT/lib/intel64" --with-real-avx512_block6-kernel-only  --with-complex-avx512_block1-kernel-only --enable-single-precision
+    - /home/elpa/wait_until_midnight.sh
+    - make -j 8
+    - export OMP_NUM_THREADS=1
+    - export LD_LIBRARY_PATH=$MKL_HOME/lib/intel64:$LD_LIBRARY_PATH
+    -  make check TEST_FLAGS='1000 500 128'
+
+
 intel-set-kernel-via-environment-variable-mpi-openmp-job:
   tags:
     - cpu

Makefile.am

@@ -204,9 +204,9 @@ endif
 
 if WITH_REAL_AVX512_BLOCK6_KERNEL
   libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx512_6hv_double_precision.c
-#if WANT_SINGLE_PRECISION_REAL
-#  libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx512_6hv_single_precision.c
-#endif
+if WANT_SINGLE_PRECISION_REAL
+  libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx512_6hv_single_precision.c
+endif
 endif
 
 

src/elpa2_kernels/elpa2_kernels_real_avx512_4hv_single_precision.c

@@ -495,7 +495,7 @@ __forceinline void hh_trafo_kernel_64_AVX512_4hv_single(float* q, float* hh, int
                 q3 = _mm512_NFMA_ps(x3, h1, q3);
                 q3 = _mm512_NFMA_ps(y3, h2, q3);
                 q3 = _mm512_NFMA_ps(z3, h3, q3);
-                q3 = _mm512_NFMA_pd(w3, h4, q3);
+                q3 = _mm512_NFMA_ps(w3, h4, q3);
 		_mm512_store_ps(&q[(i*ldq)+32],q3);
 
 		q4 = _mm512_load_ps(&q[(i*ldq)+48]);
@@ -932,7 +932,7 @@ __forceinline void hh_trafo_kernel_48_AVX512_4hv_single(float* q, float* hh, int
                 q3 = _mm512_NFMA_ps(x3, h1, q3);
                 q3 = _mm512_NFMA_ps(y3, h2, q3);
                 q3 = _mm512_NFMA_ps(z3, h3, q3);
-                q3 = _mm512_NFMA_pd(w3, h4, q3);
+                q3 = _mm512_NFMA_ps(w3, h4, q3);
 		_mm512_store_ps(&q[(i*ldq)+32],q3);
 
 //		q4 = _mm512_load_ps(&q[(i*ldq)+48]);
@@ -1369,7 +1369,7 @@ __forceinline void hh_trafo_kernel_32_AVX512_4hv_single(float* q, float* hh, int
  //               q3 = _mm512_NFMA_ps(x3, h1, q3);
 //                q3 = _mm512_NFMA_ps(y3, h2, q3);
  //               q3 = _mm512_NFMA_ps(z3, h3, q3);
- //               q3 = _mm512_NFMA_pd(w3, h4, q3);
+ //               q3 = _mm512_NFMA_ps(w3, h4, q3);
 //		_mm512_store_ps(&q[(i*ldq)+32],q3);
 
 //		q4 = _mm512_load_ps(&q[(i*ldq)+48]);
@@ -1806,7 +1806,7 @@ __forceinline void hh_trafo_kernel_16_AVX512_4hv_single(float* q, float* hh, int
  //               q3 = _mm512_NFMA_ps(x3, h1, q3);
 //                q3 = _mm512_NFMA_ps(y3, h2, q3);
  //               q3 = _mm512_NFMA_ps(z3, h3, q3);
- //               q3 = _mm512_NFMA_pd(w3, h4, q3);
+ //               q3 = _mm512_NFMA_ps(w3, h4, q3);
 //		_mm512_store_ps(&q[(i*ldq)+32],q3);
 
 //		q4 = _mm512_load_ps(&q[(i*ldq)+48]);