optimized getrf_single.c for POWER8
authorWerner Saar <wernsaar@googlemail.com>
Tue, 17 May 2016 14:19:53 +0000 (16:19 +0200)
committerWerner Saar <wernsaar@googlemail.com>
Tue, 17 May 2016 14:19:53 +0000 (16:19 +0200)
lapack/getrf/getrf_single.c

index e60a16c..9f0f36b 100644 (file)
@@ -77,10 +77,17 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
   blocking = (mn / 2 + GEMM_UNROLL_N - 1) & ~(GEMM_UNROLL_N - 1);
   if (blocking > GEMM_Q) blocking = GEMM_Q;
 
+#ifdef POWER8
+  if (blocking <= GEMM_UNROLL_N) {
+    info = GETF2(args, NULL, range_n, sa, sb, 0);
+    return info;
+  }
+#else
   if (blocking <= GEMM_UNROLL_N * 2) {
     info = GETF2(args, NULL, range_n, sa, sb, 0);
     return info;
   }
+#endif
 
   sbb = (FLOAT *)((((BLASULONG)(sb + blocking * blocking * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);