added optimized zgemv_n kernel for haswell

diff --git a/kernel/x86_64/KERNEL.HASWELL b/kernel/x86_64/KERNEL.HASWELL
index cd280e4..d126eb6 100644 (file)
--- a/kernel/x86_64/KERNEL.HASWELL
+++ b/kernel/x86_64/KERNEL.HASWELL
@@ -2,6 +2,7 @@ SGEMVNKERNEL = sgemv_n.c
 SGEMVTKERNEL = sgemv_t.c
 DGEMVNKERNEL = dgemv_n.c
 DGEMVTKERNEL = dgemv_t.c
+ZGEMVNKERNEL = zgemv_n.c
 
 SGEMMKERNEL    =  sgemm_kernel_16x4_haswell.S
 SGEMMINCOPY    =  ../generic/gemm_ncopy_16.c

diff --git a/kernel/x86_64/dgemv_n_microk_haswell-2.c b/kernel/x86_64/dgemv_n_microk_haswell-2.c
index 88c7d41..b9f462c 100644 (file)
--- a/kernel/x86_64/dgemv_n_microk_haswell-2.c
+++ b/kernel/x86_64/dgemv_n_microk_haswell-2.c
@@ -43,7 +43,7 @@ static void dgemv_kernel_16x4( BLASLONG n, float **ap, float *x, float *y)
 
        ".align 16                               \n\t"
        ".L01LOOP%=:                             \n\t"
-       "prefetcht0      192(%4,%0,8)            \n\t"
+       "prefetcht0      192(%3,%0,8)            \n\t"
        "vmovups        (%3,%0,8), %%ymm4        \n\t"  // 4 * y
        "vmovups      32(%3,%0,8), %%ymm5        \n\t"  // 4 * y
 

diff --git a/kernel/x86_64/zgemv_n.c b/kernel/x86_64/zgemv_n.c
index be5b08d..141cb35 100644 (file)
--- a/kernel/x86_64/zgemv_n.c
+++ b/kernel/x86_64/zgemv_n.c
@@ -28,6 +28,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
+#if defined(HASWELL)
+#include "zgemv_n_microk_haswell-2.c"
+#endif
+
 
 #define NBMAX 1024
 

diff --git a/kernel/x86_64/zgemv_n_microk_haswell-2.c b/kernel/x86_64/zgemv_n_microk_haswell-2.c
new file mode 100644 (file)
index 0000000..8583f96
--- /dev/null
+++ b/kernel/x86_64/zgemv_n_microk_haswell-2.c
@@ -0,0 +1,149 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define HAVE_KERNEL_16x4 1
+static void zgemv_kernel_16x4( BLASLONG n, float **ap, float *x, float *y) __attribute__ ((noinline));
+
+static void zgemv_kernel_16x4( BLASLONG n, float **ap, float *x, float *y)
+{
+
+       BLASLONG register i = 0;
+
+       __asm__  __volatile__
+       (
+       "vzeroupper                      \n\t"
+
+       "vbroadcastsd     (%2), %%ymm0                  \n\t"  // real part x0
+       "vbroadcastsd    8(%2), %%ymm1                  \n\t"  // imag part x0
+       "vbroadcastsd   16(%2), %%ymm2                  \n\t"  // real part x1
+       "vbroadcastsd   24(%2), %%ymm3                  \n\t"  // imag part x1
+       "vbroadcastsd   32(%2), %%ymm4                  \n\t"  // real part x2
+       "vbroadcastsd   40(%2), %%ymm5                  \n\t"  // imag part x2
+       "vbroadcastsd   48(%2), %%ymm6                  \n\t"  // real part x3
+       "vbroadcastsd   56(%2), %%ymm7                  \n\t"  // imag part x3
+
+
+       ".align 16                                      \n\t"
+       ".L01LOOP%=:                                    \n\t"
+       "prefetcht0      192(%4,%0,8)                   \n\t"
+       "vmovups        (%4,%0,8), %%ymm8               \n\t" // 2 complex values form a0
+       "vmovups      32(%4,%0,8), %%ymm9               \n\t" // 2 complex values form a0
+
+       "vxorpd         %%ymm12, %%ymm12, %%ymm12       \n\t"
+       "vxorpd         %%ymm13, %%ymm13, %%ymm13       \n\t"
+       "vxorpd         %%ymm14, %%ymm14, %%ymm14       \n\t"
+       "vxorpd         %%ymm15, %%ymm15, %%ymm15       \n\t"
+
+       "prefetcht0      192(%5,%0,8)                   \n\t"
+       "vmovups        (%5,%0,8), %%ymm10              \n\t" // 2 complex values form a1
+       "vmovups      32(%5,%0,8), %%ymm11              \n\t" // 2 complex values form a1
+
+       "vfmadd231pd      %%ymm8 , %%ymm0, %%ymm12      \n\t" // a_r[0] * x_r , a_i[0] * x_r, a_r[1] * x_r, a_i[1] * x_r
+       "vfmadd231pd      %%ymm8 , %%ymm1, %%ymm13      \n\t" // a_r[0] * x_i , a_i[0] * x_i, a_r[1] * x_i, a_i[1] * x_i
+       "vfmadd231pd      %%ymm9 , %%ymm0, %%ymm14      \n\t" // a_r[2] * x_r , a_i[2] * x_r, a_r[3] * x_r, a_i[3] * x_r
+       "vfmadd231pd      %%ymm9 , %%ymm1, %%ymm15      \n\t" // a_r[2] * x_i , a_i[2] * x_i, a_r[3] * x_i, a_i[3] * x_i
+
+       "prefetcht0      192(%6,%0,8)                   \n\t"
+       "vmovups        (%6,%0,8), %%ymm8               \n\t" // 2 complex values form a2
+       "vmovups      32(%6,%0,8), %%ymm9               \n\t" // 2 complex values form a2
+
+       "vfmadd231pd      %%ymm10, %%ymm2, %%ymm12      \n\t" // a_r[0] * x_r , a_i[0] * x_r, a_r[1] * x_r, a_i[1] * x_r
+       "vfmadd231pd      %%ymm10, %%ymm3, %%ymm13      \n\t" // a_r[0] * x_i , a_i[0] * x_i, a_r[1] * x_i, a_i[1] * x_i
+       "vfmadd231pd      %%ymm11, %%ymm2, %%ymm14      \n\t" // a_r[2] * x_r , a_i[2] * x_r, a_r[3] * x_r, a_i[3] * x_r
+       "vfmadd231pd      %%ymm11, %%ymm3, %%ymm15      \n\t" // a_r[2] * x_i , a_i[2] * x_i, a_r[3] * x_i, a_i[3] * x_i
+
+       "prefetcht0      192(%7,%0,8)                   \n\t"
+       "vmovups        (%7,%0,8), %%ymm10              \n\t" // 2 complex values form a3
+       "vmovups      32(%7,%0,8), %%ymm11              \n\t" // 2 complex values form a3
+
+       "vfmadd231pd      %%ymm8 , %%ymm4, %%ymm12      \n\t" // a_r[0] * x_r , a_i[0] * x_r, a_r[1] * x_r, a_i[1] * x_r
+       "vfmadd231pd      %%ymm8 , %%ymm5, %%ymm13      \n\t" // a_r[0] * x_i , a_i[0] * x_i, a_r[1] * x_i, a_i[1] * x_i
+       "vfmadd231pd      %%ymm9 , %%ymm4, %%ymm14      \n\t" // a_r[2] * x_r , a_i[2] * x_r, a_r[3] * x_r, a_i[3] * x_r
+       "vfmadd231pd      %%ymm9 , %%ymm5, %%ymm15      \n\t" // a_r[2] * x_i , a_i[2] * x_i, a_r[3] * x_i, a_i[3] * x_i
+
+       "vfmadd231pd      %%ymm10, %%ymm6, %%ymm12      \n\t" // a_r[0] * x_r , a_i[0] * x_r, a_r[1] * x_r, a_i[1] * x_r
+       "vfmadd231pd      %%ymm10, %%ymm7, %%ymm13      \n\t" // a_r[0] * x_i , a_i[0] * x_i, a_r[1] * x_i, a_i[1] * x_i
+       "vfmadd231pd      %%ymm11, %%ymm6, %%ymm14      \n\t" // a_r[2] * x_r , a_i[2] * x_r, a_r[3] * x_r, a_i[3] * x_r
+       "vfmadd231pd      %%ymm11, %%ymm7, %%ymm15      \n\t" // a_r[2] * x_i , a_i[2] * x_i, a_r[3] * x_i, a_i[3] * x_i
+
+
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+        "vpermilpd      $0x5 , %%ymm13, %%ymm13               \n\t"
+        "vpermilpd      $0x5 , %%ymm15, %%ymm15               \n\t"
+        "vaddsubpd      %%ymm13, %%ymm12, %%ymm8              \n\t"
+        "vaddsubpd      %%ymm15, %%ymm14, %%ymm9              \n\t"
+#else
+        "vpermilpd      $0x5 , %%ymm12, %%ymm12               \n\t"
+        "vpermilpd      $0x5 , %%ymm14, %%ymm14               \n\t"
+        "vaddsubpd      %%ymm13, %%ymm12, %%ymm8              \n\t"
+        "vaddsubpd      %%ymm15, %%ymm14, %%ymm9              \n\t"
+        "vpermilpd      $0x5 , %%ymm8 , %%ymm8                \n\t"
+        "vpermilpd      $0x5 , %%ymm9 , %%ymm9                \n\t"
+#endif
+
+       "prefetcht0      192(%3,%0,8)                   \n\t"
+       "vmovups          (%3,%0,8),  %%ymm12           \n\t"
+       "vmovups        32(%3,%0,8),  %%ymm13           \n\t"
+
+#if !defined(XCONJ)
+        "vaddpd         %%ymm8, %%ymm12, %%ymm12              \n\t"
+        "vaddpd         %%ymm9, %%ymm13, %%ymm13              \n\t"
+#else
+        "vaddsubpd              %%ymm12, %%ymm8, %%ymm12              \n\t"
+        "vaddsubpd              %%ymm13, %%ymm9, %%ymm13              \n\t"
+#endif
+
+
+       "vmovups  %%ymm12,   (%3,%0,8)                  \n\t" // 2 complex values to y  
+       "vmovups  %%ymm13, 32(%3,%0,8)                  \n\t"   
+
+        "addq          $8 , %0                         \n\t"
+       "subq           $4 , %1                         \n\t"           
+       "jnz            .L01LOOP%=                      \n\t"
+       "vzeroupper                      \n\t"
+
+       :
+        : 
+          "r" (i),     // 0    
+         "r" (n),      // 1
+          "r" (x),      // 2
+          "r" (y),      // 3
+          "r" (ap[0]),  // 4
+          "r" (ap[1]),  // 5
+          "r" (ap[2]),  // 6
+          "r" (ap[3])   // 7
+       : "cc", 
+         "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+         "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+         "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+         "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+         "memory"
+       );
+
+} 
+
+