added sgemv_t microkernel for haswell

diff --git a/kernel/x86_64/KERNEL.HASWELL b/kernel/x86_64/KERNEL.HASWELL
index ae316cf..288e395 100644 (file)
--- a/kernel/x86_64/KERNEL.HASWELL
+++ b/kernel/x86_64/KERNEL.HASWELL
@@ -1,3 +1,11 @@
+ifdef OS_WINDOWS
+#SGEMVNKERNEL = ../arm/gemv_n.c
+SGEMVTKERNEL = ../arm/gemv_t.c
+else
+#SGEMVNKERNEL = sgemv_n_avx.c
+SGEMVTKERNEL = sgemv_t_avx.c
+endif
+
 
 SGEMMKERNEL    =  sgemm_kernel_16x4_haswell.S
 SGEMMINCOPY    =  ../generic/gemm_ncopy_16.c

diff --git a/kernel/x86_64/sgemv_t_avx.c b/kernel/x86_64/sgemv_t_avx.c
index 7a9efa3..55fb3d6 100644 (file)
--- a/kernel/x86_64/sgemv_t_avx.c
+++ b/kernel/x86_64/sgemv_t_avx.c
@@ -30,6 +30,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #if defined(BULLDOZER) || defined(PILEDRIVER)
 #include "sgemv_t_microk_bulldozer.c"
+#elif defined(HASWELL)
+#include "sgemv_t_microk_haswell.c"
 #else
 #include "sgemv_t_microk_sandy.c"
 #endif

diff --git a/kernel/x86_64/sgemv_t_microk_haswell.c b/kernel/x86_64/sgemv_t_microk_haswell.c
new file mode 100644 (file)
index 0000000..ecb9845
--- /dev/null
+++ b/kernel/x86_64/sgemv_t_microk_haswell.c
@@ -0,0 +1,100 @@
+/***************************************************************************
+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.
+*****************************************************************************/
+
+static void  sgemv_kernel_16( long n, float alpha, float *a, long lda, float *x, float *y)
+{
+
+       //n = n / 16;
+
+       __asm__ __volatile__
+       (
+       "movq           %0,      %%rax\n\t"             // n -> rax
+       "vmovss         %1,      %%xmm1\n\t"            // alpha -> xmm1
+       "movq           %2,      %%rsi\n\t"             // adress of a -> rsi
+       "movq           %3,      %%rcx\n\t"             // value of lda > rcx
+       "movq           %4,      %%rdi\n\t"             // adress of x -> rdi
+       "movq           %5,      %%rdx\n\t"             // adress of y -> rdx
+
+       "leaq   (, %%rcx,4), %%rcx       \n\t"          // scale lda by size of float
+       "leaq   (%%rsi,%%rcx,1), %%r8    \n\t"          // pointer to next line
+
+       "vxorps         %%xmm12, %%xmm12, %%xmm12\n\t"  // set to zero
+       "vxorps         %%xmm13, %%xmm13, %%xmm13\n\t"  // set to zero
+       "vxorps         %%xmm14, %%xmm14, %%xmm14\n\t"  // set to zero
+       "vxorps         %%xmm15, %%xmm15, %%xmm15\n\t"  // set to zero
+
+       "sarq           $4, %%rax                \n\t"  // n = n / 16
+
+       ".align 16                               \n\t"
+       ".L01LOOP%=:                             \n\t"
+       // "prefetcht0  512(%%rsi)               \n\t"
+       "prefetcht0     (%%r8)                   \n\t" //prefetch next line of a
+       "vmovups        (%%rsi), %%xmm4          \n\t"
+       "vmovups     4*4(%%rsi), %%xmm5          \n\t"
+       "vmovups     8*4(%%rsi), %%xmm6          \n\t"
+       "vmovups    12*4(%%rsi), %%xmm7          \n\t"
+
+       "vfmadd231ps   0*4(%%rdi), %%xmm4, %%xmm12\n\t" // multiply a and c and add to temp
+       "vfmadd231ps   4*4(%%rdi), %%xmm5, %%xmm13\n\t" // multiply a and c and add to temp
+       "vfmadd231ps   8*4(%%rdi), %%xmm6, %%xmm14\n\t" // multiply a and c and add to temp
+       "vfmadd231ps  12*4(%%rdi), %%xmm7, %%xmm15\n\t" // multiply a and c and add to temp
+
+        "addq          $16*4    ,   %%r8        \n\t"  // increment prefetch pointer 
+        "addq          $16*4    ,   %%rsi       \n\t"  // increment pointer of a 
+        "addq          $16*4    ,   %%rdi       \n\t"  // increment pointer of c 
+       "dec            %%rax                    \n\t"  // n = n -1
+       "jnz            .L01LOOP%=               \n\t"
+
+       "vaddps         %%xmm12, %%xmm14, %%xmm12\n\t"  
+       "vaddps         %%xmm13, %%xmm15, %%xmm13\n\t"  
+       "vaddps         %%xmm12, %%xmm13, %%xmm12\n\t"  
+       "vhaddps        %%xmm12, %%xmm12, %%xmm12\n\t"  
+       "vhaddps        %%xmm12, %%xmm12, %%xmm12\n\t"  
+
+       "vmulss         %%xmm12, %%xmm1, %%xmm12\n\t"
+       "vaddss         (%%rdx), %%xmm12,%%xmm12\n\t"
+       "vmovss         %%xmm12, (%%rdx)         \n\t"  // store temp -> y
+
+       :
+        :
+          "m" (n),     // 0    
+         "m" (alpha),  // 1
+         "m" (a),      // 2
+          "m" (lda),    // 3
+          "m" (x),      // 4
+          "m" (y)       // 5
+       : "%rax", "%rcx", "%rdx", "%rsi", "%rdi", "%r8",
+         "%xmm0", "%xmm1", 
+         "%xmm4", "%xmm5", "%xmm6", "%xmm7",
+         "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+         "memory"
+       );
+
+} 
+
+
+