Merge pull request #550 from wernsaar/develop

added optimized ssymv kernels for haswell and sandybridge

kernel/x86_64/KERNEL.HASWELL

@@ -10,6 +10,8 @@ ZGEMVTKERNEL = zgemv_t_4.c
 CGEMVNKERNEL = cgemv_n_4.c
 CGEMVTKERNEL = cgemv_t_4.c
 
+SSYMV_L_KERNEL = ssymv_L.c
+SSYMV_U_KERNEL = ssymv_U.c
 DSYMV_L_KERNEL = dsymv_L.c
 DSYMV_U_KERNEL = dsymv_U.c
 

kernel/x86_64/KERNEL.SANDYBRIDGE

@@ -8,10 +8,11 @@ DDOTKERNEL = ddot.c
 CDOTKERNEL = cdot.c
 ZDOTKERNEL = zdot.c
 
+SSYMV_L_KERNEL = ssymv_L.c
+SSYMV_U_KERNEL = ssymv_U.c
 DSYMV_L_KERNEL = dsymv_L.c
 DSYMV_U_KERNEL = dsymv_U.c
 
-
 SAXPYKERNEL = saxpy.c
 DAXPYKERNEL = daxpy.c
 CAXPYKERNEL = caxpy.c

kernel/x86_64/ssymv_L.c

@@ -32,6 +32,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "ssymv_L_microk_bulldozer-2.c"
 #elif defined(NEHALEM)
 #include "ssymv_L_microk_nehalem-2.c"
+#elif defined(HASWELL)
+#include "ssymv_L_microk_haswell-2.c"
+#elif defined(SANDYBRIDGE)
+#include "ssymv_L_microk_sandy-2.c"
 #endif
 
 

kernel/x86_64/ssymv_L_microk_haswell-2.c

@@ -0,0 +1,124 @@
+/***************************************************************************
+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_4x4 1
+static void ssymv_kernel_4x4( BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FLOAT *y, FLOAT *temp1, FLOAT *temp2) __attribute__ ((noinline));
+
+static void ssymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FLOAT *y, FLOAT *temp1, FLOAT *temp2)
+{
+
+	__asm__  __volatile__
+	(
+	"vzeroupper				     \n\t"
+	"vxorps		%%xmm0 , %%xmm0 , %%xmm0     \n\t"	// temp2[0]
+	"vxorps		%%xmm1 , %%xmm1 , %%xmm1     \n\t"	// temp2[1]
+	"vxorps		%%xmm2 , %%xmm2 , %%xmm2     \n\t"	// temp2[2]
+	"vxorps		%%xmm3 , %%xmm3 , %%xmm3     \n\t"	// temp2[3]
+	"vbroadcastss   (%8),    %%xmm4	             \n\t"	// temp1[0]
+	"vbroadcastss  4(%8),    %%xmm5	             \n\t"	// temp1[1]
+	"vbroadcastss  8(%8),    %%xmm6	             \n\t"	// temp1[1]
+	"vbroadcastss 12(%8),    %%xmm7	             \n\t"	// temp1[1]
+
+	".align 16				     \n\t"
+	"1:				     \n\t"
+
+	"vmovups	(%3,%0,4), %%xmm9	           \n\t"  // 2 * y
+	"vmovups	(%2,%0,4), %%xmm8	           \n\t"  // 2 * x
+
+	"vmovups	(%4,%0,4), %%xmm12	           \n\t"  // 2 * a
+	"vmovups	(%5,%0,4), %%xmm13	           \n\t"  // 2 * a
+	"vmovups	(%6,%0,4), %%xmm14	           \n\t"  // 2 * a
+	"vmovups	(%7,%0,4), %%xmm15	           \n\t"  // 2 * a
+
+	"vfmadd231ps	%%xmm4, %%xmm12 , %%xmm9  \n\t"  // y     += temp1 * a
+	"vfmadd231ps	%%xmm8, %%xmm12 , %%xmm0  \n\t"  // temp2 += x * a
+
+	"vfmadd231ps	%%xmm5, %%xmm13 , %%xmm9  \n\t"  // y     += temp1 * a
+	"vfmadd231ps	%%xmm8, %%xmm13 , %%xmm1  \n\t"  // temp2 += x * a
+
+	"vfmadd231ps	%%xmm6, %%xmm14 , %%xmm9  \n\t"  // y     += temp1 * a
+	"vfmadd231ps	%%xmm8, %%xmm14 , %%xmm2  \n\t"  // temp2 += x * a
+
+	"vfmadd231ps	%%xmm7, %%xmm15 , %%xmm9  \n\t"  // y     += temp1 * a
+	"vfmadd231ps	%%xmm8, %%xmm15 , %%xmm3  \n\t"  // temp2 += x * a
+
+	"vmovups	%%xmm9 ,  (%3,%0,4)		   \n\t"
+
+	"addq		$4 , %0	  	 	      \n\t"
+	"cmpq		%0 , %1			      \n\t"
+	"jnz		1b		      \n\t"
+
+	"vmovss		  (%9), %%xmm4		      \n\t"
+	"vmovss		 4(%9), %%xmm5		      \n\t"
+	"vmovss		 8(%9), %%xmm6		      \n\t"
+	"vmovss		12(%9), %%xmm7		      \n\t"
+
+	"vhaddps        %%xmm0, %%xmm0, %%xmm0  \n\t"
+	"vhaddps        %%xmm1, %%xmm1, %%xmm1  \n\t"
+	"vhaddps        %%xmm2, %%xmm2, %%xmm2  \n\t"
+	"vhaddps        %%xmm3, %%xmm3, %%xmm3  \n\t"
+
+	"vhaddps        %%xmm0, %%xmm0, %%xmm0  \n\t"
+	"vhaddps        %%xmm1, %%xmm1, %%xmm1  \n\t"
+	"vhaddps        %%xmm2, %%xmm2, %%xmm2  \n\t"
+	"vhaddps        %%xmm3, %%xmm3, %%xmm3  \n\t"
+
+	"vaddss		%%xmm4, %%xmm0, %%xmm0  \n\t"
+	"vaddss		%%xmm5, %%xmm1, %%xmm1  \n\t"
+	"vaddss		%%xmm6, %%xmm2, %%xmm2  \n\t"
+	"vaddss		%%xmm7, %%xmm3, %%xmm3  \n\t"
+
+	"vmovss         %%xmm0 ,  (%9)		\n\t"	// save temp2
+	"vmovss         %%xmm1 , 4(%9)		\n\t"	// save temp2
+	"vmovss         %%xmm2 , 8(%9)		\n\t"	// save temp2
+	"vmovss         %%xmm3 ,12(%9)		\n\t"	// save temp2
+	"vzeroupper				     \n\t"
+
+	:
+        : 
+          "r" (from),	// 0	
+	  "r" (to),  	// 1
+          "r" (x),      // 2
+          "r" (y),      // 3
+          "r" (a[0]),	// 4
+          "r" (a[1]),	// 5
+          "r" (a[2]),	// 6
+          "r" (a[3]),	// 8
+          "r" (temp1),  // 8
+          "r" (temp2)   // 9
+	: "cc", 
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+
+} 
+
+

kernel/x86_64/ssymv_L_microk_sandy-2.c

@@ -0,0 +1,243 @@
+/***************************************************************************
+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_4x4 1
+static void ssymv_kernel_4x4( BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FLOAT *y, FLOAT *temp1, FLOAT *temp2) __attribute__ ((noinline));
+
+static void ssymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FLOAT *y, FLOAT *temp1, FLOAT *temp2)
+{
+
+  if ( ( to - from ) & 4 )
+  {
+	__asm__  __volatile__
+	(
+	"vzeroupper				     \n\t"
+	"vxorps		%%xmm0 , %%xmm0 , %%xmm0     \n\t"	// temp2[0]
+	"vxorps		%%xmm1 , %%xmm1 , %%xmm1     \n\t"	// temp2[1]
+	"vxorps		%%xmm2 , %%xmm2 , %%xmm2     \n\t"	// temp2[2]
+	"vxorps		%%xmm3 , %%xmm3 , %%xmm3     \n\t"	// temp2[3]
+	"vbroadcastss   (%8),    %%xmm4	             \n\t"	// temp1[0]
+	"vbroadcastss  4(%8),    %%xmm5	             \n\t"	// temp1[1]
+	"vbroadcastss  8(%8),    %%xmm6	             \n\t"	// temp1[1]
+	"vbroadcastss 12(%8),    %%xmm7	             \n\t"	// temp1[1]
+
+	".align 16				     \n\t"
+	"1:				     \n\t"
+
+	"vmovups	(%3,%0,4), %%xmm9	           \n\t"  // 2 * y
+	"vmovups	(%2,%0,4), %%xmm8	           \n\t"  // 2 * x
+
+	"vmovups	(%4,%0,4), %%xmm12	           \n\t"  // 2 * a
+	"vmovups	(%5,%0,4), %%xmm13	           \n\t"  // 2 * a
+	"vmovups	(%6,%0,4), %%xmm14	           \n\t"  // 2 * a
+	"vmovups	(%7,%0,4), %%xmm15	           \n\t"  // 2 * a
+
+        "vmulps         %%xmm4, %%xmm12 , %%xmm10          \n\t"
+        "vmulps         %%xmm8, %%xmm12 , %%xmm11          \n\t"
+        "vaddps         %%xmm9, %%xmm10 , %%xmm9           \n\t"
+        "vaddps         %%xmm0, %%xmm11 , %%xmm0           \n\t"
+
+        "vmulps         %%xmm5, %%xmm13 , %%xmm10          \n\t"
+        "vmulps         %%xmm8, %%xmm13 , %%xmm11          \n\t"
+        "vaddps         %%xmm9, %%xmm10 , %%xmm9           \n\t"
+        "vaddps         %%xmm1, %%xmm11 , %%xmm1           \n\t"
+
+        "vmulps         %%xmm6, %%xmm14 , %%xmm10          \n\t"
+        "vmulps         %%xmm8, %%xmm14 , %%xmm11          \n\t"
+        "vaddps         %%xmm9, %%xmm10 , %%xmm9           \n\t"
+        "vaddps         %%xmm2, %%xmm11 , %%xmm2           \n\t"
+
+        "vmulps         %%xmm7, %%xmm15 , %%xmm10          \n\t"
+        "vmulps         %%xmm8, %%xmm15 , %%xmm11          \n\t"
+        "vaddps         %%xmm9, %%xmm10 , %%xmm9           \n\t"
+        "vaddps         %%xmm3, %%xmm11 , %%xmm3           \n\t"
+
+	"vmovups	%%xmm9 ,  (%3,%0,4)		   \n\t"
+
+	"addq		$4 , %0	  	 	      \n\t"
+	"cmpq		%0 , %1			      \n\t"
+	"jnz		1b		      \n\t"
+
+	"vmovss		  (%9), %%xmm4		      \n\t"
+	"vmovss		 4(%9), %%xmm5		      \n\t"
+	"vmovss		 8(%9), %%xmm6		      \n\t"
+	"vmovss		12(%9), %%xmm7		      \n\t"
+
+	"vhaddps        %%xmm0, %%xmm0, %%xmm0  \n\t"
+	"vhaddps        %%xmm1, %%xmm1, %%xmm1  \n\t"
+	"vhaddps        %%xmm2, %%xmm2, %%xmm2  \n\t"
+	"vhaddps        %%xmm3, %%xmm3, %%xmm3  \n\t"
+
+	"vhaddps        %%xmm0, %%xmm0, %%xmm0  \n\t"
+	"vhaddps        %%xmm1, %%xmm1, %%xmm1  \n\t"
+	"vhaddps        %%xmm2, %%xmm2, %%xmm2  \n\t"
+	"vhaddps        %%xmm3, %%xmm3, %%xmm3  \n\t"
+
+	"vaddss		%%xmm4, %%xmm0, %%xmm0  \n\t"
+	"vaddss		%%xmm5, %%xmm1, %%xmm1  \n\t"
+	"vaddss		%%xmm6, %%xmm2, %%xmm2  \n\t"
+	"vaddss		%%xmm7, %%xmm3, %%xmm3  \n\t"
+
+	"vmovss         %%xmm0 ,  (%9)		\n\t"	// save temp2
+	"vmovss         %%xmm1 , 4(%9)		\n\t"	// save temp2
+	"vmovss         %%xmm2 , 8(%9)		\n\t"	// save temp2
+	"vmovss         %%xmm3 ,12(%9)		\n\t"	// save temp2
+	"vzeroupper				     \n\t"
+
+	:
+        : 
+          "r" (from),	// 0	
+	  "r" (to),  	// 1
+          "r" (x),      // 2
+          "r" (y),      // 3
+          "r" (a[0]),	// 4
+          "r" (a[1]),	// 5
+          "r" (a[2]),	// 6
+          "r" (a[3]),	// 8
+          "r" (temp1),  // 8
+          "r" (temp2)   // 9
+	: "cc", 
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+	return;
+  }
+
+
+	__asm__  __volatile__
+	(
+	"vzeroupper				     \n\t"
+	"vxorps		%%ymm0 , %%ymm0 , %%ymm0     \n\t"	// temp2[0]
+	"vxorps		%%ymm1 , %%ymm1 , %%ymm1     \n\t"	// temp2[1]
+	"vxorps		%%ymm2 , %%ymm2 , %%ymm2     \n\t"	// temp2[2]
+	"vxorps		%%ymm3 , %%ymm3 , %%ymm3     \n\t"	// temp2[3]
+	"vbroadcastss   (%8),    %%ymm4	             \n\t"	// temp1[0]
+	"vbroadcastss  4(%8),    %%ymm5	             \n\t"	// temp1[1]
+	"vbroadcastss  8(%8),    %%ymm6	             \n\t"	// temp1[1]
+	"vbroadcastss 12(%8),    %%ymm7	             \n\t"	// temp1[1]
+
+	".align 16				     \n\t"
+	"1:				     \n\t"
+
+	"vmovups	(%3,%0,4), %%ymm9	           \n\t"  // 2 * y
+	"vmovups	(%2,%0,4), %%ymm8	           \n\t"  // 2 * x
+
+	"vmovups	(%4,%0,4), %%ymm12	           \n\t"  // 2 * a
+	"vmovups	(%5,%0,4), %%ymm13	           \n\t"  // 2 * a
+	"vmovups	(%6,%0,4), %%ymm14	           \n\t"  // 2 * a
+	"vmovups	(%7,%0,4), %%ymm15	           \n\t"  // 2 * a
+
+        "vmulps         %%ymm4, %%ymm12 , %%ymm10          \n\t"
+        "vmulps         %%ymm8, %%ymm12 , %%ymm11          \n\t"
+        "vaddps         %%ymm9, %%ymm10 , %%ymm9           \n\t"
+        "vaddps         %%ymm0, %%ymm11 , %%ymm0           \n\t"
+
+        "vmulps         %%ymm5, %%ymm13 , %%ymm10          \n\t"
+        "vmulps         %%ymm8, %%ymm13 , %%ymm11          \n\t"
+        "vaddps         %%ymm9, %%ymm10 , %%ymm9           \n\t"
+        "vaddps         %%ymm1, %%ymm11 , %%ymm1           \n\t"
+
+        "vmulps         %%ymm6, %%ymm14 , %%ymm10          \n\t"
+        "vmulps         %%ymm8, %%ymm14 , %%ymm11          \n\t"
+        "vaddps         %%ymm9, %%ymm10 , %%ymm9           \n\t"
+        "vaddps         %%ymm2, %%ymm11 , %%ymm2           \n\t"
+
+        "vmulps         %%ymm7, %%ymm15 , %%ymm10          \n\t"
+        "vmulps         %%ymm8, %%ymm15 , %%ymm11          \n\t"
+        "vaddps         %%ymm9, %%ymm10 , %%ymm9           \n\t"
+        "vaddps         %%ymm3, %%ymm11 , %%ymm3           \n\t"
+
+	"vmovups	%%ymm9 ,  (%3,%0,4)		   \n\t"
+
+	"addq		$8 , %0	  	 	      \n\t"
+	"cmpq		%0 , %1			      \n\t"
+	"jnz		1b		      \n\t"
+
+	"vmovss		  (%9), %%xmm4		      \n\t"
+	"vmovss		 4(%9), %%xmm5		      \n\t"
+	"vmovss		 8(%9), %%xmm6		      \n\t"
+	"vmovss		12(%9), %%xmm7		      \n\t"
+
+        "vextractf128 $0x01, %%ymm0 , %%xmm12         \n\t"
+        "vextractf128 $0x01, %%ymm1 , %%xmm13         \n\t"
+        "vextractf128 $0x01, %%ymm2 , %%xmm14         \n\t"
+        "vextractf128 $0x01, %%ymm3 , %%xmm15         \n\t"
+
+        "vaddps         %%xmm0, %%xmm12, %%xmm0       \n\t"
+        "vaddps         %%xmm1, %%xmm13, %%xmm1       \n\t"
+        "vaddps         %%xmm2, %%xmm14, %%xmm2       \n\t"
+        "vaddps         %%xmm3, %%xmm15, %%xmm3       \n\t"
+
+	"vhaddps        %%xmm0, %%xmm0, %%xmm0  \n\t"
+	"vhaddps        %%xmm1, %%xmm1, %%xmm1  \n\t"
+	"vhaddps        %%xmm2, %%xmm2, %%xmm2  \n\t"
+	"vhaddps        %%xmm3, %%xmm3, %%xmm3  \n\t"
+
+	"vhaddps        %%xmm0, %%xmm0, %%xmm0  \n\t"
+	"vhaddps        %%xmm1, %%xmm1, %%xmm1  \n\t"
+	"vhaddps        %%xmm2, %%xmm2, %%xmm2  \n\t"
+	"vhaddps        %%xmm3, %%xmm3, %%xmm3  \n\t"
+
+	"vaddss		%%xmm4, %%xmm0, %%xmm0  \n\t"
+	"vaddss		%%xmm5, %%xmm1, %%xmm1  \n\t"
+	"vaddss		%%xmm6, %%xmm2, %%xmm2  \n\t"
+	"vaddss		%%xmm7, %%xmm3, %%xmm3  \n\t"
+
+	"vmovss         %%xmm0 ,  (%9)		\n\t"	// save temp2
+	"vmovss         %%xmm1 , 4(%9)		\n\t"	// save temp2
+	"vmovss         %%xmm2 , 8(%9)		\n\t"	// save temp2
+	"vmovss         %%xmm3 ,12(%9)		\n\t"	// save temp2
+	"vzeroupper				     \n\t"
+
+	:
+        : 
+          "r" (from),	// 0	
+	  "r" (to),  	// 1
+          "r" (x),      // 2
+          "r" (y),      // 3
+          "r" (a[0]),	// 4
+          "r" (a[1]),	// 5
+          "r" (a[2]),	// 6
+          "r" (a[3]),	// 8
+          "r" (temp1),  // 8
+          "r" (temp2)   // 9
+	: "cc", 
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+
+
+} 
+
+

kernel/x86_64/ssymv_U.c

@@ -33,6 +33,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "ssymv_U_microk_bulldozer-2.c"
 #elif defined(NEHALEM)
 #include "ssymv_U_microk_nehalem-2.c"
+#elif defined(HASWELL)
+#include "ssymv_U_microk_haswell-2.c"
+#elif defined(SANDYBRIDGE)
+#include "ssymv_U_microk_sandy-2.c"
 #endif
 
 #ifndef HAVE_KERNEL_4x4

kernel/x86_64/ssymv_U_microk_haswell-2.c

@@ -0,0 +1,136 @@
+/***************************************************************************
+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_4x4 1
+static void ssymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT *a3, FLOAT *x, FLOAT *y, FLOAT *temp1, FLOAT *temp2) __attribute__ ((noinline));
+
+static void ssymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT *a3, FLOAT *x, FLOAT *y, FLOAT *temp1, FLOAT *temp2)
+{
+
+	BLASLONG register i = 0;
+
+	__asm__  __volatile__
+	(
+	"vzeroupper				     \n\t"
+	"vxorps		%%ymm0 , %%ymm0 , %%ymm0     \n\t"	// temp2[0]
+	"vxorps		%%ymm1 , %%ymm1 , %%ymm1     \n\t"	// temp2[1]
+	"vxorps		%%ymm2 , %%ymm2 , %%ymm2     \n\t"	// temp2[2]
+	"vxorps		%%ymm3 , %%ymm3 , %%ymm3     \n\t"	// temp2[3]
+	"vbroadcastss   (%8),    %%ymm4	             \n\t"	// temp1[0]
+	"vbroadcastss  4(%8),    %%ymm5	             \n\t"	// temp1[1]
+	"vbroadcastss  8(%8),    %%ymm6	             \n\t"	// temp1[1]
+	"vbroadcastss 12(%8),    %%ymm7	             \n\t"	// temp1[1]
+	"xorq           %0,%0                        \n\t"
+
+	".align 16				     \n\t"
+	"1:				     \n\t"
+
+	"vmovups	(%3,%0,4), %%ymm9	           \n\t"  // 2 * y
+	"vmovups	(%2,%0,4), %%ymm8	           \n\t"  // 2 * x
+
+	"vmovups	(%4,%0,4), %%ymm12	           \n\t"  // 2 * a
+	"vmovups	(%5,%0,4), %%ymm13	           \n\t"  // 2 * a
+	"vmovups	(%6,%0,4), %%ymm14	           \n\t"  // 2 * a
+	"vmovups	(%7,%0,4), %%ymm15	           \n\t"  // 2 * a
+
+	"vfmadd231ps	%%ymm4, %%ymm12 , %%ymm9  \n\t"  // y     += temp1 * a
+	"vfmadd231ps	%%ymm8, %%ymm12 , %%ymm0  \n\t"  // temp2 += x * a
+
+	"vfmadd231ps	%%ymm5, %%ymm13 , %%ymm9  \n\t"  // y     += temp1 * a
+	"vfmadd231ps	%%ymm8, %%ymm13 , %%ymm1  \n\t"  // temp2 += x * a
+
+	"vfmadd231ps	%%ymm6, %%ymm14 , %%ymm9  \n\t"  // y     += temp1 * a
+	"vfmadd231ps	%%ymm8, %%ymm14 , %%ymm2  \n\t"  // temp2 += x * a
+
+	"vfmadd231ps	%%ymm7, %%ymm15 , %%ymm9  \n\t"  // y     += temp1 * a
+	"vfmadd231ps	%%ymm8, %%ymm15 , %%ymm3  \n\t"  // temp2 += x * a
+
+	"vmovups	%%ymm9 ,  (%3,%0,4)	      \n\t"
+	"addq		$8 , %0	  	 	      \n\t"
+	"subq		$8 , %1	  	 	      \n\t"
+
+	"jnz		1b		      \n\t"
+
+	"vmovss		  (%9), %%xmm4		      \n\t"
+	"vmovss		 4(%9), %%xmm5		      \n\t"
+	"vmovss		 8(%9), %%xmm6		      \n\t"
+	"vmovss		12(%9), %%xmm7		      \n\t"
+
+	"vextractf128 $0x01, %%ymm0 , %%xmm12	      \n\t"
+	"vextractf128 $0x01, %%ymm1 , %%xmm13	      \n\t"
+	"vextractf128 $0x01, %%ymm2 , %%xmm14	      \n\t"
+	"vextractf128 $0x01, %%ymm3 , %%xmm15	      \n\t"
+
+	"vaddps	        %%xmm0, %%xmm12, %%xmm0	      \n\t"
+	"vaddps	        %%xmm1, %%xmm13, %%xmm1	      \n\t"
+	"vaddps	        %%xmm2, %%xmm14, %%xmm2	      \n\t"
+	"vaddps	        %%xmm3, %%xmm15, %%xmm3	      \n\t"
+
+	"vhaddps        %%xmm0, %%xmm0, %%xmm0  \n\t"
+	"vhaddps        %%xmm1, %%xmm1, %%xmm1  \n\t"
+	"vhaddps        %%xmm2, %%xmm2, %%xmm2  \n\t"
+	"vhaddps        %%xmm3, %%xmm3, %%xmm3  \n\t"
+
+	"vhaddps        %%xmm0, %%xmm0, %%xmm0  \n\t"
+	"vhaddps        %%xmm1, %%xmm1, %%xmm1  \n\t"
+	"vhaddps        %%xmm2, %%xmm2, %%xmm2  \n\t"
+	"vhaddps        %%xmm3, %%xmm3, %%xmm3  \n\t"
+
+	"vaddsd		%%xmm4, %%xmm0, %%xmm0  \n\t"
+	"vaddsd		%%xmm5, %%xmm1, %%xmm1  \n\t"
+	"vaddsd		%%xmm6, %%xmm2, %%xmm2  \n\t"
+	"vaddsd		%%xmm7, %%xmm3, %%xmm3  \n\t"
+
+	"vmovss         %%xmm0 ,  (%9)		\n\t"	// save temp2
+	"vmovss         %%xmm1 , 4(%9)		\n\t"	// save temp2
+	"vmovss         %%xmm2 , 8(%9)		\n\t"	// save temp2
+	"vmovss         %%xmm3 ,12(%9)		\n\t"	// save temp2
+	"vzeroupper				     \n\t"
+
+	:
+        : 
+          "r" (i),	// 0	
+	  "r" (n),  	// 1
+          "r" (x),      // 2
+          "r" (y),      // 3
+          "r" (a0),	// 4
+          "r" (a1),	// 5
+          "r" (a2),	// 6
+          "r" (a3),	// 8
+          "r" (temp1),  // 8
+          "r" (temp2)   // 9
+	: "cc", 
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+

kernel/x86_64/ssymv_U_microk_sandy-2.c

@@ -0,0 +1,144 @@
+/***************************************************************************
+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_4x4 1
+static void ssymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT *a3, FLOAT *x, FLOAT *y, FLOAT *temp1, FLOAT *temp2) __attribute__ ((noinline));
+
+static void ssymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT *a3, FLOAT *x, FLOAT *y, FLOAT *temp1, FLOAT *temp2)
+{
+
+	BLASLONG register i = 0;
+
+	__asm__  __volatile__
+	(
+	"vzeroupper				     \n\t"
+	"vxorps		%%ymm0 , %%ymm0 , %%ymm0     \n\t"	// temp2[0]
+	"vxorps		%%ymm1 , %%ymm1 , %%ymm1     \n\t"	// temp2[1]
+	"vxorps		%%ymm2 , %%ymm2 , %%ymm2     \n\t"	// temp2[2]
+	"vxorps		%%ymm3 , %%ymm3 , %%ymm3     \n\t"	// temp2[3]
+	"vbroadcastss   (%8),    %%ymm4	             \n\t"	// temp1[0]
+	"vbroadcastss  4(%8),    %%ymm5	             \n\t"	// temp1[1]
+	"vbroadcastss  8(%8),    %%ymm6	             \n\t"	// temp1[1]
+	"vbroadcastss 12(%8),    %%ymm7	             \n\t"	// temp1[1]
+	"xorq           %0,%0                        \n\t"
+
+	".align 16				     \n\t"
+	"1:				     \n\t"
+
+	"vmovups	(%3,%0,4), %%ymm9	           \n\t"  // 2 * y
+	"vmovups	(%2,%0,4), %%ymm8	           \n\t"  // 2 * x
+
+	"vmovups	(%4,%0,4), %%ymm12	           \n\t"  // 2 * a
+	"vmovups	(%5,%0,4), %%ymm13	           \n\t"  // 2 * a
+	"vmovups	(%6,%0,4), %%ymm14	           \n\t"  // 2 * a
+	"vmovups	(%7,%0,4), %%ymm15	           \n\t"  // 2 * a
+
+	"vmulps		%%ymm4, %%ymm12 , %%ymm10	   \n\t"
+	"vaddps		%%ymm9, %%ymm10 , %%ymm9	   \n\t"
+	"vmulps		%%ymm8, %%ymm12 , %%ymm11	   \n\t"
+	"vaddps		%%ymm0, %%ymm11 , %%ymm0	   \n\t"
+
+	"vmulps		%%ymm5, %%ymm13 , %%ymm10	   \n\t"
+	"vaddps		%%ymm9, %%ymm10 , %%ymm9	   \n\t"
+	"vmulps		%%ymm8, %%ymm13 , %%ymm11	   \n\t"
+	"vaddps		%%ymm1, %%ymm11 , %%ymm1	   \n\t"
+
+	"vmulps		%%ymm6, %%ymm14 , %%ymm10	   \n\t"
+	"vaddps		%%ymm9, %%ymm10 , %%ymm9	   \n\t"
+	"vmulps		%%ymm8, %%ymm14 , %%ymm11	   \n\t"
+	"vaddps		%%ymm2, %%ymm11 , %%ymm2	   \n\t"
+
+	"vmulps		%%ymm7, %%ymm15 , %%ymm10	   \n\t"
+	"vaddps		%%ymm9, %%ymm10 , %%ymm9	   \n\t"
+	"vmulps		%%ymm8, %%ymm15 , %%ymm11	   \n\t"
+	"vaddps		%%ymm3, %%ymm11 , %%ymm3	   \n\t"
+
+	"vmovups	%%ymm9 ,  (%3,%0,4)	      \n\t"
+	"addq		$8 , %0	  	 	      \n\t"
+	"subq		$8 , %1	  	 	      \n\t"
+
+	"jnz		1b		      \n\t"
+
+	"vmovss		  (%9), %%xmm4		      \n\t"
+	"vmovss		 4(%9), %%xmm5		      \n\t"
+	"vmovss		 8(%9), %%xmm6		      \n\t"
+	"vmovss		12(%9), %%xmm7		      \n\t"
+
+	"vextractf128 $0x01, %%ymm0 , %%xmm12	      \n\t"
+	"vextractf128 $0x01, %%ymm1 , %%xmm13	      \n\t"
+	"vextractf128 $0x01, %%ymm2 , %%xmm14	      \n\t"
+	"vextractf128 $0x01, %%ymm3 , %%xmm15	      \n\t"
+
+	"vaddps	        %%xmm0, %%xmm12, %%xmm0	      \n\t"
+	"vaddps	        %%xmm1, %%xmm13, %%xmm1	      \n\t"
+	"vaddps	        %%xmm2, %%xmm14, %%xmm2	      \n\t"
+	"vaddps	        %%xmm3, %%xmm15, %%xmm3	      \n\t"
+
+	"vhaddps        %%xmm0, %%xmm0, %%xmm0  \n\t"
+	"vhaddps        %%xmm1, %%xmm1, %%xmm1  \n\t"
+	"vhaddps        %%xmm2, %%xmm2, %%xmm2  \n\t"
+	"vhaddps        %%xmm3, %%xmm3, %%xmm3  \n\t"
+
+	"vhaddps        %%xmm0, %%xmm0, %%xmm0  \n\t"
+	"vhaddps        %%xmm1, %%xmm1, %%xmm1  \n\t"
+	"vhaddps        %%xmm2, %%xmm2, %%xmm2  \n\t"
+	"vhaddps        %%xmm3, %%xmm3, %%xmm3  \n\t"
+
+	"vaddsd		%%xmm4, %%xmm0, %%xmm0  \n\t"
+	"vaddsd		%%xmm5, %%xmm1, %%xmm1  \n\t"
+	"vaddsd		%%xmm6, %%xmm2, %%xmm2  \n\t"
+	"vaddsd		%%xmm7, %%xmm3, %%xmm3  \n\t"
+
+	"vmovss         %%xmm0 ,  (%9)		\n\t"	// save temp2
+	"vmovss         %%xmm1 , 4(%9)		\n\t"	// save temp2
+	"vmovss         %%xmm2 , 8(%9)		\n\t"	// save temp2
+	"vmovss         %%xmm3 ,12(%9)		\n\t"	// save temp2
+	"vzeroupper				     \n\t"
+
+	:
+        : 
+          "r" (i),	// 0	
+	  "r" (n),  	// 1
+          "r" (x),      // 2
+          "r" (y),      // 3
+          "r" (a0),	// 4
+          "r" (a1),	// 5
+          "r" (a2),	// 6
+          "r" (a3),	// 8
+          "r" (temp1),  // 8
+          "r" (temp2)   // 9
+	: "cc", 
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+