From 5c511639720d428904cf2c010f550fce090acc12 Mon Sep 17 00:00:00 2001 From: Werner Saar Date: Thu, 9 Apr 2015 09:45:23 +0200 Subject: [PATCH] added optimized cdot- and zdot-kernel for steamroller --- kernel/x86_64/KERNEL.STEAMROLLER | 2 +- kernel/x86_64/cdot.c | 4 +- kernel/x86_64/cdot_microk_steamroller-2.c | 196 ++++++++++++++++++++++++++++++ kernel/x86_64/zdot.c | 4 +- kernel/x86_64/zdot_microk_steamroller-2.c | 193 +++++++++++++++++++++++++++++ 5 files changed, 396 insertions(+), 3 deletions(-) create mode 100644 kernel/x86_64/cdot_microk_steamroller-2.c create mode 100644 kernel/x86_64/zdot_microk_steamroller-2.c diff --git a/kernel/x86_64/KERNEL.STEAMROLLER b/kernel/x86_64/KERNEL.STEAMROLLER index 8926010..fbe04ca 100644 --- a/kernel/x86_64/KERNEL.STEAMROLLER +++ b/kernel/x86_64/KERNEL.STEAMROLLER @@ -3,7 +3,7 @@ CAXPYKERNEL = caxpy.c ZAXPYKERNEL = zaxpy.c SDOTKERNEL = sdot.c -DDOTKERNEL = ddot_bullozer.S +DDOTKERNEL = ddot.c CDOTKERNEL = cdot.c ZDOTKERNEL = zdot.c diff --git a/kernel/x86_64/cdot.c b/kernel/x86_64/cdot.c index 1e9e320..bfe7073 100644 --- a/kernel/x86_64/cdot.c +++ b/kernel/x86_64/cdot.c @@ -30,8 +30,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. #include -#if defined(BULLDOZER) || defined(PILEDRIVER) || defined(STEAMROLLER) +#if defined(BULLDOZER) || defined(PILEDRIVER) #include "cdot_microk_bulldozer-2.c" +#elif defined(STEAMROLLER) +#include "cdot_microk_steamroller-2.c" #elif defined(NEHALEM) #include "cdot_microk_nehalem-2.c" #elif defined(HASWELL) diff --git a/kernel/x86_64/cdot_microk_steamroller-2.c b/kernel/x86_64/cdot_microk_steamroller-2.c new file mode 100644 index 0000000..76a3aa0 --- /dev/null +++ b/kernel/x86_64/cdot_microk_steamroller-2.c @@ -0,0 +1,196 @@ +/*************************************************************************** +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_16 1 +static void cdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y , FLOAT *dot) __attribute__ ((noinline)); + +static void cdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot) +{ + + + BLASLONG register i = 0; + + if ( n < 1280 ) + { + + __asm__ __volatile__ + ( + "vzeroupper \n\t" + "vxorps %%xmm0, %%xmm0, %%xmm0 \n\t" + "vxorps %%xmm1, %%xmm1, %%xmm1 \n\t" + "vxorps %%xmm2, %%xmm2, %%xmm2 \n\t" + "vxorps %%xmm3, %%xmm3, %%xmm3 \n\t" + "vxorps %%xmm4, %%xmm4, %%xmm4 \n\t" + "vxorps %%xmm5, %%xmm5, %%xmm5 \n\t" + "vxorps %%xmm6, %%xmm6, %%xmm6 \n\t" + "vxorps %%xmm7, %%xmm7, %%xmm7 \n\t" + + ".align 16 \n\t" + "1: \n\t" + "vmovups (%2,%0,4), %%xmm8 \n\t" // 2 * x + "vmovups 16(%2,%0,4), %%xmm9 \n\t" // 2 * x + + "vmovups (%3,%0,4), %%xmm12 \n\t" // 2 * y + "vmovups 16(%3,%0,4), %%xmm13 \n\t" // 2 * y + + "vmovups 32(%2,%0,4), %%xmm10 \n\t" // 2 * x + "vmovups 48(%2,%0,4), %%xmm11 \n\t" // 2 * x + + "vmovups 32(%3,%0,4), %%xmm14 \n\t" // 2 * y + "vmovups 48(%3,%0,4), %%xmm15 \n\t" // 2 * y + + "vfmadd231ps %%xmm8 , %%xmm12, %%xmm0 \n\t" // x_r * y_r, x_i * y_i + "vfmadd231ps %%xmm9 , %%xmm13, %%xmm1 \n\t" // x_r * y_r, x_i * y_i + + "vpermilps $0xb1 , %%xmm12, %%xmm12 \n\t" // exchange real and imag part + "vpermilps $0xb1 , %%xmm13, %%xmm13 \n\t" + + "vfmadd231ps %%xmm10, %%xmm14, %%xmm2 \n\t" // x_r * y_r, x_i * y_i + "vfmadd231ps %%xmm11, %%xmm15, %%xmm3 \n\t" // x_r * y_r, x_i * y_i + + "vpermilps $0xb1 , %%xmm14, %%xmm14 \n\t" + "vpermilps $0xb1 , %%xmm15, %%xmm15 \n\t" + + "vfmadd231ps %%xmm8 , %%xmm12, %%xmm4 \n\t" // x_r * y_i, x_i * y_r + "addq $16 , %0 \n\t" + "vfmadd231ps %%xmm9 , %%xmm13, %%xmm5 \n\t" // x_r * y_i, x_i * y_r + "vfmadd231ps %%xmm10, %%xmm14, %%xmm6 \n\t" // x_r * y_i, x_i * y_r + "subq $8 , %1 \n\t" + "vfmadd231ps %%xmm11, %%xmm15, %%xmm7 \n\t" // x_r * y_i, x_i * y_r + + "jnz 1b \n\t" + + "vaddps %%xmm0, %%xmm1, %%xmm0 \n\t" + "vaddps %%xmm2, %%xmm3, %%xmm2 \n\t" + "vaddps %%xmm0, %%xmm2, %%xmm0 \n\t" + + "vaddps %%xmm4, %%xmm5, %%xmm4 \n\t" + "vaddps %%xmm6, %%xmm7, %%xmm6 \n\t" + "vaddps %%xmm4, %%xmm6, %%xmm4 \n\t" + + "vmovups %%xmm0, (%4) \n\t" + "vmovups %%xmm4, 16(%4) \n\t" + "vzeroupper \n\t" + + : + : + "r" (i), // 0 + "r" (n), // 1 + "r" (x), // 2 + "r" (y), // 3 + "r" (dot) // 4 + : "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 %%xmm0, %%xmm0, %%xmm0 \n\t" + "vxorps %%xmm1, %%xmm1, %%xmm1 \n\t" + "vxorps %%xmm2, %%xmm2, %%xmm2 \n\t" + "vxorps %%xmm3, %%xmm3, %%xmm3 \n\t" + "vxorps %%xmm4, %%xmm4, %%xmm4 \n\t" + "vxorps %%xmm5, %%xmm5, %%xmm5 \n\t" + "vxorps %%xmm6, %%xmm6, %%xmm6 \n\t" + "vxorps %%xmm7, %%xmm7, %%xmm7 \n\t" + + ".align 16 \n\t" + "1: \n\t" + "prefetcht0 512(%2,%0,4) \n\t" + "vmovups (%2,%0,4), %%xmm8 \n\t" // 2 * x + "vmovups 16(%2,%0,4), %%xmm9 \n\t" // 2 * x + + "prefetcht0 512(%3,%0,4) \n\t" + "vmovups (%3,%0,4), %%xmm12 \n\t" // 2 * y + "vmovups 16(%3,%0,4), %%xmm13 \n\t" // 2 * y + + "vmovups 32(%2,%0,4), %%xmm10 \n\t" // 2 * x + "vmovups 48(%2,%0,4), %%xmm11 \n\t" // 2 * x + + "vmovups 32(%3,%0,4), %%xmm14 \n\t" // 2 * y + "vmovups 48(%3,%0,4), %%xmm15 \n\t" // 2 * y + + "vfmadd231ps %%xmm8 , %%xmm12, %%xmm0 \n\t" // x_r * y_r, x_i * y_i + "vfmadd231ps %%xmm9 , %%xmm13, %%xmm1 \n\t" // x_r * y_r, x_i * y_i + + "vpermilps $0xb1 , %%xmm12, %%xmm12 \n\t" // exchange real and imag part + "vpermilps $0xb1 , %%xmm13, %%xmm13 \n\t" + + "vfmadd231ps %%xmm10, %%xmm14, %%xmm2 \n\t" // x_r * y_r, x_i * y_i + "vfmadd231ps %%xmm11, %%xmm15, %%xmm3 \n\t" // x_r * y_r, x_i * y_i + + "vpermilps $0xb1 , %%xmm14, %%xmm14 \n\t" + "vpermilps $0xb1 , %%xmm15, %%xmm15 \n\t" + + "vfmadd231ps %%xmm8 , %%xmm12, %%xmm4 \n\t" // x_r * y_i, x_i * y_r + "addq $16 , %0 \n\t" + "vfmadd231ps %%xmm9 , %%xmm13, %%xmm5 \n\t" // x_r * y_i, x_i * y_r + "vfmadd231ps %%xmm10, %%xmm14, %%xmm6 \n\t" // x_r * y_i, x_i * y_r + "subq $8 , %1 \n\t" + "vfmadd231ps %%xmm11, %%xmm15, %%xmm7 \n\t" // x_r * y_i, x_i * y_r + + "jnz 1b \n\t" + + "vaddps %%xmm0, %%xmm1, %%xmm0 \n\t" + "vaddps %%xmm2, %%xmm3, %%xmm2 \n\t" + "vaddps %%xmm0, %%xmm2, %%xmm0 \n\t" + + "vaddps %%xmm4, %%xmm5, %%xmm4 \n\t" + "vaddps %%xmm6, %%xmm7, %%xmm6 \n\t" + "vaddps %%xmm4, %%xmm6, %%xmm4 \n\t" + + "vmovups %%xmm0, (%4) \n\t" + "vmovups %%xmm4, 16(%4) \n\t" + "vzeroupper \n\t" + + : + : + "r" (i), // 0 + "r" (n), // 1 + "r" (x), // 2 + "r" (y), // 3 + "r" (dot) // 4 + : "cc", + "%xmm0", "%xmm1", "%xmm2", "%xmm3", + "%xmm4", "%xmm5", "%xmm6", "%xmm7", + "%xmm8", "%xmm9", "%xmm10", "%xmm11", + "%xmm12", "%xmm13", "%xmm14", "%xmm15", + "memory" + ); + + +} + + diff --git a/kernel/x86_64/zdot.c b/kernel/x86_64/zdot.c index e13858e..e11b62c 100644 --- a/kernel/x86_64/zdot.c +++ b/kernel/x86_64/zdot.c @@ -30,8 +30,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. #include -#if defined(BULLDOZER) || defined(PILEDRIVER) || defined(STEAMROLLER) +#if defined(BULLDOZER) || defined(PILEDRIVER) #include "zdot_microk_bulldozer-2.c" +#elif defined(STEAMROLLER) +#include "zdot_microk_steamroller-2.c" #elif defined(NEHALEM) #include "zdot_microk_nehalem-2.c" #elif defined(HASWELL) diff --git a/kernel/x86_64/zdot_microk_steamroller-2.c b/kernel/x86_64/zdot_microk_steamroller-2.c new file mode 100644 index 0000000..325f74a --- /dev/null +++ b/kernel/x86_64/zdot_microk_steamroller-2.c @@ -0,0 +1,193 @@ +/*************************************************************************** +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_8 1 +static void zdot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y , FLOAT *dot) __attribute__ ((noinline)); + +static void zdot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot) +{ + + + BLASLONG register i = 0; + + if ( n < 640 ) + { + + __asm__ __volatile__ + ( + "vzeroupper \n\t" + "vxorpd %%xmm0, %%xmm0, %%xmm0 \n\t" + "vxorpd %%xmm1, %%xmm1, %%xmm1 \n\t" + "vxorpd %%xmm2, %%xmm2, %%xmm2 \n\t" + "vxorpd %%xmm3, %%xmm3, %%xmm3 \n\t" + "vxorpd %%xmm4, %%xmm4, %%xmm4 \n\t" + "vxorpd %%xmm5, %%xmm5, %%xmm5 \n\t" + "vxorpd %%xmm6, %%xmm6, %%xmm6 \n\t" + "vxorpd %%xmm7, %%xmm7, %%xmm7 \n\t" + + ".align 16 \n\t" + "1: \n\t" + //"prefetcht0 512(%2,%0,8) \n\t" + "vmovups (%2,%0,8), %%xmm8 \n\t" // 1 * x + "vmovups 16(%2,%0,8), %%xmm9 \n\t" // 1 * x + + // "prefetcht0 512(%3,%0,8) \n\t" + "vmovups (%3,%0,8), %%xmm12 \n\t" // 1 * y + "vmovups 16(%3,%0,8), %%xmm13 \n\t" // 1 * y + + "vmovups 32(%2,%0,8), %%xmm10 \n\t" // 1 * x + "vmovups 48(%2,%0,8), %%xmm11 \n\t" // 1 * x + + "vmovups 32(%3,%0,8), %%xmm14 \n\t" // 1 * y + "vmovups 48(%3,%0,8), %%xmm15 \n\t" // 1 * y + + "vfmadd231pd %%xmm8 , %%xmm12, %%xmm0 \n\t" // x_r * y_r, x_i * y_i + "vfmadd231pd %%xmm9 , %%xmm13, %%xmm1 \n\t" // x_r * y_r, x_i * y_i + "vpermilpd $0x1 , %%xmm13, %%xmm13 \n\t" + "vpermilpd $0x1 , %%xmm12, %%xmm12 \n\t" + "vfmadd231pd %%xmm10, %%xmm14, %%xmm2 \n\t" // x_r * y_r, x_i * y_i + "vfmadd231pd %%xmm11, %%xmm15, %%xmm3 \n\t" // x_r * y_r, x_i * y_i + "vpermilpd $0x1 , %%xmm14, %%xmm14 \n\t" + "vpermilpd $0x1 , %%xmm15, %%xmm15 \n\t" + + "vfmadd231pd %%xmm8 , %%xmm12, %%xmm4 \n\t" // x_r * y_i, x_i * y_r + "addq $8 , %0 \n\t" + "vfmadd231pd %%xmm9 , %%xmm13, %%xmm5 \n\t" // x_r * y_i, x_i * y_r + "vfmadd231pd %%xmm10, %%xmm14, %%xmm6 \n\t" // x_r * y_i, x_i * y_r + "subq $4 , %1 \n\t" + "vfmadd231pd %%xmm11, %%xmm15, %%xmm7 \n\t" // x_r * y_i, x_i * y_r + + "jnz 1b \n\t" + + "vaddpd %%xmm0, %%xmm1, %%xmm0 \n\t" + "vaddpd %%xmm2, %%xmm3, %%xmm2 \n\t" + "vaddpd %%xmm0, %%xmm2, %%xmm0 \n\t" + + "vaddpd %%xmm4, %%xmm5, %%xmm4 \n\t" + "vaddpd %%xmm6, %%xmm7, %%xmm6 \n\t" + "vaddpd %%xmm4, %%xmm6, %%xmm4 \n\t" + + "vmovups %%xmm0, (%4) \n\t" + "vmovups %%xmm4, 16(%4) \n\t" + "vzeroupper \n\t" + + : + : + "r" (i), // 0 + "r" (n), // 1 + "r" (x), // 2 + "r" (y), // 3 + "r" (dot) // 4 + : "cc", + "%xmm0", "%xmm1", "%xmm2", "%xmm3", + "%xmm4", "%xmm5", "%xmm6", "%xmm7", + "%xmm8", "%xmm9", "%xmm10", "%xmm11", + "%xmm12", "%xmm13", "%xmm14", "%xmm15", + "memory" + ); + + return; + } + + + __asm__ __volatile__ + ( + "vzeroupper \n\t" + "vxorpd %%xmm0, %%xmm0, %%xmm0 \n\t" + "vxorpd %%xmm1, %%xmm1, %%xmm1 \n\t" + "vxorpd %%xmm2, %%xmm2, %%xmm2 \n\t" + "vxorpd %%xmm3, %%xmm3, %%xmm3 \n\t" + "vxorpd %%xmm4, %%xmm4, %%xmm4 \n\t" + "vxorpd %%xmm5, %%xmm5, %%xmm5 \n\t" + "vxorpd %%xmm6, %%xmm6, %%xmm6 \n\t" + "vxorpd %%xmm7, %%xmm7, %%xmm7 \n\t" + + ".align 16 \n\t" + "1: \n\t" + "prefetcht0 512(%2,%0,8) \n\t" + "vmovups (%2,%0,8), %%xmm8 \n\t" // 1 * x + "vmovups 16(%2,%0,8), %%xmm9 \n\t" // 1 * x + + "prefetcht0 512(%3,%0,8) \n\t" + "vmovups (%3,%0,8), %%xmm12 \n\t" // 1 * y + "vmovups 16(%3,%0,8), %%xmm13 \n\t" // 1 * y + + "vmovups 32(%2,%0,8), %%xmm10 \n\t" // 1 * x + "vmovups 48(%2,%0,8), %%xmm11 \n\t" // 1 * x + + "vmovups 32(%3,%0,8), %%xmm14 \n\t" // 1 * y + "vmovups 48(%3,%0,8), %%xmm15 \n\t" // 1 * y + + "vfmadd231pd %%xmm8 , %%xmm12, %%xmm0 \n\t" // x_r * y_r, x_i * y_i + "vfmadd231pd %%xmm9 , %%xmm13, %%xmm1 \n\t" // x_r * y_r, x_i * y_i + "vpermilpd $0x1 , %%xmm13, %%xmm13 \n\t" + "vpermilpd $0x1 , %%xmm12, %%xmm12 \n\t" + "vfmadd231pd %%xmm10, %%xmm14, %%xmm2 \n\t" // x_r * y_r, x_i * y_i + "vfmadd231pd %%xmm11, %%xmm15, %%xmm3 \n\t" // x_r * y_r, x_i * y_i + "vpermilpd $0x1 , %%xmm14, %%xmm14 \n\t" + "vpermilpd $0x1 , %%xmm15, %%xmm15 \n\t" + + "vfmadd231pd %%xmm8 , %%xmm12, %%xmm4 \n\t" // x_r * y_i, x_i * y_r + "addq $8 , %0 \n\t" + "vfmadd231pd %%xmm9 , %%xmm13, %%xmm5 \n\t" // x_r * y_i, x_i * y_r + "vfmadd231pd %%xmm10, %%xmm14, %%xmm6 \n\t" // x_r * y_i, x_i * y_r + "subq $4 , %1 \n\t" + "vfmadd231pd %%xmm11, %%xmm15, %%xmm7 \n\t" // x_r * y_i, x_i * y_r + + "jnz 1b \n\t" + + "vaddpd %%xmm0, %%xmm1, %%xmm0 \n\t" + "vaddpd %%xmm2, %%xmm3, %%xmm2 \n\t" + "vaddpd %%xmm0, %%xmm2, %%xmm0 \n\t" + + "vaddpd %%xmm4, %%xmm5, %%xmm4 \n\t" + "vaddpd %%xmm6, %%xmm7, %%xmm6 \n\t" + "vaddpd %%xmm4, %%xmm6, %%xmm4 \n\t" + + "vmovups %%xmm0, (%4) \n\t" + "vmovups %%xmm4, 16(%4) \n\t" + "vzeroupper \n\t" + + : + : + "r" (i), // 0 + "r" (n), // 1 + "r" (x), // 2 + "r" (y), // 3 + "r" (dot) // 4 + : "cc", + "%xmm0", "%xmm1", "%xmm2", "%xmm3", + "%xmm4", "%xmm5", "%xmm6", "%xmm7", + "%xmm8", "%xmm9", "%xmm10", "%xmm11", + "%xmm12", "%xmm13", "%xmm14", "%xmm15", + "memory" + ); + + +} + + -- 2.7.4