From 24f58c8bb15530a3fb00ed43d187e7d2f0cd2fc7 Mon Sep 17 00:00:00 2001 From: Werner Saar Date: Mon, 18 May 2015 12:40:07 +0200 Subject: [PATCH] added optimized cscal and zscal kernels for steamroller --- kernel/x86_64/cscal.c | 4 +- kernel/x86_64/cscal_microk_steamroller-2.c | 349 +++++++++++++++++++++ kernel/x86_64/zscal.c | 4 +- kernel/x86_64/zscal_microk_steamroller-2.c | 349 +++++++++++++++++++++ 4 files changed, 704 insertions(+), 2 deletions(-) create mode 100644 kernel/x86_64/cscal_microk_steamroller-2.c create mode 100644 kernel/x86_64/zscal_microk_steamroller-2.c diff --git a/kernel/x86_64/cscal.c b/kernel/x86_64/cscal.c index 8c8eba420..5d86b1929 100644 --- a/kernel/x86_64/cscal.c +++ b/kernel/x86_64/cscal.c @@ -30,8 +30,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. #if defined(HASWELL) #include "cscal_microk_haswell-2.c" -#elif defined(BULLDOZER) || defined(PILEDRIVER) || defined(STEAMROLLER) +#elif defined(BULLDOZER) || defined(PILEDRIVER) #include "cscal_microk_bulldozer-2.c" +#elif defined(STEAMROLLER) +#include "cscal_microk_steamroller-2.c" #elif defined(SANDYBRIDGE) #include "cscal_microk_bulldozer-2.c" #endif diff --git a/kernel/x86_64/cscal_microk_steamroller-2.c b/kernel/x86_64/cscal_microk_steamroller-2.c new file mode 100644 index 000000000..763e7add4 --- /dev/null +++ b/kernel/x86_64/cscal_microk_steamroller-2.c @@ -0,0 +1,349 @@ +/*************************************************************************** +Copyright (c) 2014-2015, 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 cscal_kernel_16( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline)); + +static void cscal_kernel_16( BLASLONG n, FLOAT *alpha, FLOAT *x) +{ + + + __asm__ __volatile__ + ( + "vbroadcastss (%2), %%ymm0 \n\t" // da_r + "vbroadcastss 4(%2), %%ymm1 \n\t" // da_i + + "addq $128, %1 \n\t" + + "vmovups -128(%1), %%ymm4 \n\t" + "vmovups -96(%1), %%ymm5 \n\t" + "vmovups -64(%1), %%ymm6 \n\t" + "vmovups -32(%1), %%ymm7 \n\t" + + "vpermilps $0xb1 , %%ymm4, %%ymm12 \n\t" + "vpermilps $0xb1 , %%ymm5, %%ymm13 \n\t" + "vpermilps $0xb1 , %%ymm6, %%ymm14 \n\t" + "vpermilps $0xb1 , %%ymm7, %%ymm15 \n\t" + + "subq $16, %0 \n\t" + "jz 2f \n\t" + + ".align 16 \n\t" + "1: \n\t" + + "prefetcht0 512(%1) \n\t" + // ".align 2 \n\t" + + "vmulps %%ymm0, %%ymm4 , %%ymm8 \n\t" // da_r*x0 , da_r *x1 + "vmovups 0(%1), %%ymm4 \n\t" + "vmulps %%ymm0, %%ymm5 , %%ymm9 \n\t" + "vmovups 32(%1), %%ymm5 \n\t" + "prefetcht0 768(%1) \n\t" + "vmulps %%ymm0, %%ymm6 , %%ymm10 \n\t" + "vmovups 64(%1), %%ymm6 \n\t" + "vmulps %%ymm0, %%ymm7 , %%ymm11 \n\t" + "vmovups 96(%1), %%ymm7 \n\t" + + "vmulps %%ymm1, %%ymm12, %%ymm12 \n\t" // da_i*x1 , da_i *x0 + "vmulps %%ymm1, %%ymm13, %%ymm13 \n\t" + "vaddsubps %%ymm12 , %%ymm8 , %%ymm8 \n\t" + "vaddsubps %%ymm13 , %%ymm9 , %%ymm9 \n\t" + "vmulps %%ymm1, %%ymm14, %%ymm14 \n\t" + "vmulps %%ymm1, %%ymm15, %%ymm15 \n\t" + "vaddsubps %%ymm14 , %%ymm10, %%ymm10 \n\t" + "vaddsubps %%ymm15 , %%ymm11, %%ymm11 \n\t" + + "vmovups %%ymm8 , -128(%1) \n\t" + "vmovups %%ymm9 , -96(%1) \n\t" + "vmovups %%ymm10, -64(%1) \n\t" + "vmovups %%ymm11, -32(%1) \n\t" + "vpermilps $0xb1 , %%ymm4, %%ymm12 \n\t" + "vpermilps $0xb1 , %%ymm5, %%ymm13 \n\t" + "vpermilps $0xb1 , %%ymm6, %%ymm14 \n\t" + "vpermilps $0xb1 , %%ymm7, %%ymm15 \n\t" + + "addq $128 ,%1 \n\t" + "subq $16, %0 \n\t" + "jnz 1b \n\t" + + "2: \n\t" + + + "vmulps %%ymm0, %%ymm4 , %%ymm8 \n\t" // da_r*x0 , da_r *x1 + "vmulps %%ymm0, %%ymm5 , %%ymm9 \n\t" + "vmulps %%ymm0, %%ymm6 , %%ymm10 \n\t" + "vmulps %%ymm0, %%ymm7 , %%ymm11 \n\t" + + "vmulps %%ymm1, %%ymm12, %%ymm12 \n\t" // da_i*x1 , da_i *x0 + "vaddsubps %%ymm12 , %%ymm8 , %%ymm8 \n\t" + "vmulps %%ymm1, %%ymm13, %%ymm13 \n\t" + "vaddsubps %%ymm13 , %%ymm9 , %%ymm9 \n\t" + "vmulps %%ymm1, %%ymm14, %%ymm14 \n\t" + "vaddsubps %%ymm14 , %%ymm10, %%ymm10 \n\t" + "vmulps %%ymm1, %%ymm15, %%ymm15 \n\t" + "vaddsubps %%ymm15 , %%ymm11, %%ymm11 \n\t" + + "vmovups %%ymm8 , -128(%1) \n\t" + "vmovups %%ymm9 , -96(%1) \n\t" + "vmovups %%ymm10, -64(%1) \n\t" + "vmovups %%ymm11, -32(%1) \n\t" + + "vzeroupper \n\t" + + : + : + "r" (n), // 0 + "r" (x), // 1 + "r" (alpha) // 2 + : "cc", "0", "1", + "%xmm0", "%xmm1", "%xmm2", "%xmm3", + "%xmm4", "%xmm5", "%xmm6", "%xmm7", + "%xmm8", "%xmm9", "%xmm10", "%xmm11", + "%xmm12", "%xmm13", "%xmm14", "%xmm15", + "memory" + ); + +} + +static void cscal_kernel_16_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline)); + +static void cscal_kernel_16_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x) +{ + + + __asm__ __volatile__ + ( + "vxorps %%ymm0, %%ymm0, %%ymm0 \n\t" + "vbroadcastss 4(%2), %%ymm1 \n\t" // da_i + + "addq $128, %1 \n\t" + + "vmovups -128(%1), %%ymm4 \n\t" + "vmovups -96(%1), %%ymm5 \n\t" + "vmovups -64(%1), %%ymm6 \n\t" + "vmovups -32(%1), %%ymm7 \n\t" + + "vpermilps $0xb1 , %%ymm4, %%ymm12 \n\t" + "vpermilps $0xb1 , %%ymm5, %%ymm13 \n\t" + "vpermilps $0xb1 , %%ymm6, %%ymm14 \n\t" + "vpermilps $0xb1 , %%ymm7, %%ymm15 \n\t" + + "subq $16, %0 \n\t" + "jz 2f \n\t" + + ".align 16 \n\t" + "1: \n\t" + + //"prefetcht0 128(%1) \n\t" + // ".align 2 \n\t" + + "vmovups 0(%1), %%ymm4 \n\t" + "vmovups 32(%1), %%ymm5 \n\t" + "vmovups 64(%1), %%ymm6 \n\t" + "vmovups 96(%1), %%ymm7 \n\t" + + "vmulps %%ymm1, %%ymm12, %%ymm12 \n\t" // da_i*x1 , da_i *x0 + "vaddsubps %%ymm12 , %%ymm0 , %%ymm8 \n\t" + "vmulps %%ymm1, %%ymm13, %%ymm13 \n\t" + "vaddsubps %%ymm13 , %%ymm0 , %%ymm9 \n\t" + "vmulps %%ymm1, %%ymm14, %%ymm14 \n\t" + "vaddsubps %%ymm14 , %%ymm0 , %%ymm10 \n\t" + "vmulps %%ymm1, %%ymm15, %%ymm15 \n\t" + "vaddsubps %%ymm15 , %%ymm0 , %%ymm11 \n\t" + + "vmovups %%ymm8 , -128(%1) \n\t" + "vpermilps $0xb1 , %%ymm4, %%ymm12 \n\t" + "vmovups %%ymm9 , -96(%1) \n\t" + "vpermilps $0xb1 , %%ymm5, %%ymm13 \n\t" + "vmovups %%ymm10, -64(%1) \n\t" + "vpermilps $0xb1 , %%ymm6, %%ymm14 \n\t" + "vmovups %%ymm11, -32(%1) \n\t" + "vpermilps $0xb1 , %%ymm7, %%ymm15 \n\t" + + "addq $128 ,%1 \n\t" + "subq $16, %0 \n\t" + "jnz 1b \n\t" + + "2: \n\t" + + "vmulps %%ymm1, %%ymm12, %%ymm12 \n\t" // da_i*x1 , da_i *x0 + "vaddsubps %%ymm12 , %%ymm0 , %%ymm8 \n\t" + "vmulps %%ymm1, %%ymm13, %%ymm13 \n\t" + "vaddsubps %%ymm13 , %%ymm0 , %%ymm9 \n\t" + "vmulps %%ymm1, %%ymm14, %%ymm14 \n\t" + "vaddsubps %%ymm14 , %%ymm0 , %%ymm10 \n\t" + "vmulps %%ymm1, %%ymm15, %%ymm15 \n\t" + "vaddsubps %%ymm15 , %%ymm0 , %%ymm11 \n\t" + + "vmovups %%ymm8 , -128(%1) \n\t" + "vmovups %%ymm9 , -96(%1) \n\t" + "vmovups %%ymm10, -64(%1) \n\t" + "vmovups %%ymm11, -32(%1) \n\t" + + "vzeroupper \n\t" + + : + : + "r" (n), // 0 + "r" (x), // 1 + "r" (alpha) // 2 + : "cc", "0", "1", + "%xmm0", "%xmm1", "%xmm2", "%xmm3", + "%xmm4", "%xmm5", "%xmm6", "%xmm7", + "%xmm8", "%xmm9", "%xmm10", "%xmm11", + "%xmm12", "%xmm13", "%xmm14", "%xmm15", + "memory" + ); + +} + + + +static void cscal_kernel_16_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline)); + +static void cscal_kernel_16_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x) +{ + + + __asm__ __volatile__ + ( + "vbroadcastss (%2), %%ymm0 \n\t" // da_r + + "addq $128, %1 \n\t" + + "vmovups -128(%1), %%ymm4 \n\t" + "vmovups -96(%1), %%ymm5 \n\t" + "vmovups -64(%1), %%ymm6 \n\t" + "vmovups -32(%1), %%ymm7 \n\t" + + + "subq $16, %0 \n\t" + "jz 2f \n\t" + + ".align 16 \n\t" + "1: \n\t" + + //"prefetcht0 128(%1) \n\t" + // ".align 2 \n\t" + + "vmulps %%ymm0, %%ymm4 , %%ymm8 \n\t" // da_r*x0 , da_r *x1 + "vmovups 0(%1), %%ymm4 \n\t" + "vmulps %%ymm0, %%ymm5 , %%ymm9 \n\t" + "vmovups 32(%1), %%ymm5 \n\t" + "vmulps %%ymm0, %%ymm6 , %%ymm10 \n\t" + "vmovups 64(%1), %%ymm6 \n\t" + "vmulps %%ymm0, %%ymm7 , %%ymm11 \n\t" + "vmovups 96(%1), %%ymm7 \n\t" + + "vmovups %%ymm8 , -128(%1) \n\t" + "vmovups %%ymm9 , -96(%1) \n\t" + "vmovups %%ymm10, -64(%1) \n\t" + "vmovups %%ymm11, -32(%1) \n\t" + + "addq $128 ,%1 \n\t" + "subq $16, %0 \n\t" + "jnz 1b \n\t" + + "2: \n\t" + + + "vmulps %%ymm0, %%ymm4 , %%ymm8 \n\t" // da_r*x0 , da_r *x1 + "vmulps %%ymm0, %%ymm5 , %%ymm9 \n\t" + "vmulps %%ymm0, %%ymm6 , %%ymm10 \n\t" + "vmulps %%ymm0, %%ymm7 , %%ymm11 \n\t" + + "vmovups %%ymm8 , -128(%1) \n\t" + "vmovups %%ymm9 , -96(%1) \n\t" + "vmovups %%ymm10, -64(%1) \n\t" + "vmovups %%ymm11, -32(%1) \n\t" + + "vzeroupper \n\t" + + : + : + "r" (n), // 0 + "r" (x), // 1 + "r" (alpha) // 2 + : "cc", "%0", "%1", + "%xmm0", "%xmm1", "%xmm2", "%xmm3", + "%xmm4", "%xmm5", "%xmm6", "%xmm7", + "%xmm8", "%xmm9", "%xmm10", "%xmm11", + "%xmm12", "%xmm13", "%xmm14", "%xmm15", + "memory" + ); + +} + + +static void cscal_kernel_16_zero( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline)); + +static void cscal_kernel_16_zero( BLASLONG n, FLOAT *alpha, FLOAT *x) +{ + + + __asm__ __volatile__ + ( + "vxorps %%ymm0, %%ymm0, %%ymm0 \n\t" + + "addq $128, %1 \n\t" + + ".align 16 \n\t" + "1: \n\t" + + //"prefetcht0 128(%1) \n\t" + // ".align 2 \n\t" + + "vmovups %%ymm0 , -128(%1) \n\t" + "vmovups %%ymm0 , -96(%1) \n\t" + "vmovups %%ymm0 , -64(%1) \n\t" + "vmovups %%ymm0 , -32(%1) \n\t" + + "addq $128 ,%1 \n\t" + "subq $16, %0 \n\t" + "jnz 1b \n\t" + + "vzeroupper \n\t" + + : + : + "r" (n), // 0 + "r" (x), // 1 + "r" (alpha) // 2 + : "cc", "0", "1", + "%xmm0", "%xmm1", "%xmm2", "%xmm3", + "%xmm4", "%xmm5", "%xmm6", "%xmm7", + "%xmm8", "%xmm9", "%xmm10", "%xmm11", + "%xmm12", "%xmm13", "%xmm14", "%xmm15", + "memory" + ); + +} + + + diff --git a/kernel/x86_64/zscal.c b/kernel/x86_64/zscal.c index 91bd44161..a96766032 100644 --- a/kernel/x86_64/zscal.c +++ b/kernel/x86_64/zscal.c @@ -30,8 +30,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. #if defined(HASWELL) #include "zscal_microk_haswell-2.c" -#elif defined(BULLDOZER) || defined(PILEDRIVER) || defined(STEAMROLLER) +#elif defined(BULLDOZER) || defined(PILEDRIVER) #include "zscal_microk_bulldozer-2.c" +#elif defined(STEAMROLLER) +#include "zscal_microk_steamroller-2.c" #endif diff --git a/kernel/x86_64/zscal_microk_steamroller-2.c b/kernel/x86_64/zscal_microk_steamroller-2.c new file mode 100644 index 000000000..d611bf570 --- /dev/null +++ b/kernel/x86_64/zscal_microk_steamroller-2.c @@ -0,0 +1,349 @@ +/*************************************************************************** +Copyright (c) 2014-2015, 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 zscal_kernel_8( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline)); + +static void zscal_kernel_8( BLASLONG n, FLOAT *alpha, FLOAT *x) +{ + + + __asm__ __volatile__ + ( + "vbroadcastsd (%2), %%ymm0 \n\t" // da_r + "vbroadcastsd 8(%2), %%ymm1 \n\t" // da_i + + "addq $128, %1 \n\t" + + "vmovups -128(%1), %%ymm4 \n\t" + "vmovups -96(%1), %%ymm5 \n\t" + "vmovups -64(%1), %%ymm6 \n\t" + "vmovups -32(%1), %%ymm7 \n\t" + + "vpermilpd $0x05 , %%ymm4, %%ymm12 \n\t" + "vpermilpd $0x05 , %%ymm5, %%ymm13 \n\t" + "vpermilpd $0x05 , %%ymm6, %%ymm14 \n\t" + "vpermilpd $0x05 , %%ymm7, %%ymm15 \n\t" + + "subq $8 , %0 \n\t" + "jz 2f \n\t" + + ".align 16 \n\t" + "1: \n\t" + + "prefetcht0 320(%1) \n\t" + "prefetcht0 384(%1) \n\t" + // ".align 2 \n\t" + + "vmulpd %%ymm0, %%ymm4 , %%ymm8 \n\t" // da_r*x0 , da_r *x1 + "vmovups 0(%1), %%ymm4 \n\t" + "vmulpd %%ymm0, %%ymm5 , %%ymm9 \n\t" + "vmovups 32(%1), %%ymm5 \n\t" + "vmulpd %%ymm0, %%ymm6 , %%ymm10 \n\t" + "vmovups 64(%1), %%ymm6 \n\t" + "vmulpd %%ymm0, %%ymm7 , %%ymm11 \n\t" + "vmovups 96(%1), %%ymm7 \n\t" + + "vmulpd %%ymm1, %%ymm12, %%ymm12 \n\t" // da_i*x1 , da_i *x0 + "vaddsubpd %%ymm12 , %%ymm8 , %%ymm8 \n\t" + "vmulpd %%ymm1, %%ymm13, %%ymm13 \n\t" + "vaddsubpd %%ymm13 , %%ymm9 , %%ymm9 \n\t" + "vmulpd %%ymm1, %%ymm14, %%ymm14 \n\t" + "vaddsubpd %%ymm14 , %%ymm10, %%ymm10 \n\t" + "vmulpd %%ymm1, %%ymm15, %%ymm15 \n\t" + "vaddsubpd %%ymm15 , %%ymm11, %%ymm11 \n\t" + + "vmovups %%ymm8 , -128(%1) \n\t" + "vpermilpd $0x05 , %%ymm4, %%ymm12 \n\t" + "vmovups %%ymm9 , -96(%1) \n\t" + "vpermilpd $0x05 , %%ymm5, %%ymm13 \n\t" + "vmovups %%ymm10, -64(%1) \n\t" + "vpermilpd $0x05 , %%ymm6, %%ymm14 \n\t" + "vmovups %%ymm11, -32(%1) \n\t" + "vpermilpd $0x05 , %%ymm7, %%ymm15 \n\t" + + "addq $128 ,%1 \n\t" + "subq $8 , %0 \n\t" + "jnz 1b \n\t" + + "2: \n\t" + + + "vmulpd %%ymm0, %%ymm4 , %%ymm8 \n\t" // da_r*x0 , da_r *x1 + "vmulpd %%ymm0, %%ymm5 , %%ymm9 \n\t" + "vmulpd %%ymm0, %%ymm6 , %%ymm10 \n\t" + "vmulpd %%ymm0, %%ymm7 , %%ymm11 \n\t" + + "vmulpd %%ymm1, %%ymm12, %%ymm12 \n\t" // da_i*x1 , da_i *x0 + "vaddsubpd %%ymm12 , %%ymm8 , %%ymm8 \n\t" + "vmulpd %%ymm1, %%ymm13, %%ymm13 \n\t" + "vaddsubpd %%ymm13 , %%ymm9 , %%ymm9 \n\t" + "vmulpd %%ymm1, %%ymm14, %%ymm14 \n\t" + "vaddsubpd %%ymm14 , %%ymm10, %%ymm10 \n\t" + "vmulpd %%ymm1, %%ymm15, %%ymm15 \n\t" + "vaddsubpd %%ymm15 , %%ymm11, %%ymm11 \n\t" + + "vmovups %%ymm8 , -128(%1) \n\t" + "vmovups %%ymm9 , -96(%1) \n\t" + "vmovups %%ymm10, -64(%1) \n\t" + "vmovups %%ymm11, -32(%1) \n\t" + + "vzeroupper \n\t" + + : + : + "r" (n), // 0 + "r" (x), // 1 + "r" (alpha) // 2 + : "cc", "%0", "%1", + "%xmm0", "%xmm1", "%xmm2", "%xmm3", + "%xmm4", "%xmm5", "%xmm6", "%xmm7", + "%xmm8", "%xmm9", "%xmm10", "%xmm11", + "%xmm12", "%xmm13", "%xmm14", "%xmm15", + "memory" + ); + +} + +static void zscal_kernel_8_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline)); + +static void zscal_kernel_8_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x) +{ + + + __asm__ __volatile__ + ( + "vxorpd %%ymm0, %%ymm0, %%ymm0 \n\t" + "vbroadcastsd 8(%2), %%ymm1 \n\t" // da_i + + "addq $128, %1 \n\t" + + "vmovups -128(%1), %%ymm4 \n\t" + "vmovups -96(%1), %%ymm5 \n\t" + "vmovups -64(%1), %%ymm6 \n\t" + "vmovups -32(%1), %%ymm7 \n\t" + + "vpermilpd $0x05 , %%ymm4, %%ymm12 \n\t" + "vpermilpd $0x05 , %%ymm5, %%ymm13 \n\t" + "vpermilpd $0x05 , %%ymm6, %%ymm14 \n\t" + "vpermilpd $0x05 , %%ymm7, %%ymm15 \n\t" + + "subq $8 , %0 \n\t" + "jz 2f \n\t" + + ".align 16 \n\t" + "1: \n\t" + + //"prefetcht0 128(%1) \n\t" + // ".align 2 \n\t" + + "vmovups 0(%1), %%ymm4 \n\t" + "vmovups 32(%1), %%ymm5 \n\t" + "vmovups 64(%1), %%ymm6 \n\t" + "vmovups 96(%1), %%ymm7 \n\t" + + "vmulpd %%ymm1, %%ymm12, %%ymm12 \n\t" // da_i*x1 , da_i *x0 + "vaddsubpd %%ymm12 , %%ymm0 , %%ymm8 \n\t" + "vmulpd %%ymm1, %%ymm13, %%ymm13 \n\t" + "vaddsubpd %%ymm13 , %%ymm0 , %%ymm9 \n\t" + "vmulpd %%ymm1, %%ymm14, %%ymm14 \n\t" + "vaddsubpd %%ymm14 , %%ymm0 , %%ymm10 \n\t" + "vmulpd %%ymm1, %%ymm15, %%ymm15 \n\t" + "vaddsubpd %%ymm15 , %%ymm0 , %%ymm11 \n\t" + + "vmovups %%ymm8 , -128(%1) \n\t" + "vpermilpd $0x05 , %%ymm4, %%ymm12 \n\t" + "vmovups %%ymm9 , -96(%1) \n\t" + "vpermilpd $0x05 , %%ymm5, %%ymm13 \n\t" + "vmovups %%ymm10, -64(%1) \n\t" + "vpermilpd $0x05 , %%ymm6, %%ymm14 \n\t" + "vmovups %%ymm11, -32(%1) \n\t" + "vpermilpd $0x05 , %%ymm7, %%ymm15 \n\t" + + "addq $128 ,%1 \n\t" + "subq $8 , %0 \n\t" + "jnz 1b \n\t" + + "2: \n\t" + + "vmulpd %%ymm1, %%ymm12, %%ymm12 \n\t" // da_i*x1 , da_i *x0 + "vaddsubpd %%ymm12 , %%ymm0 , %%ymm8 \n\t" + "vmulpd %%ymm1, %%ymm13, %%ymm13 \n\t" + "vaddsubpd %%ymm13 , %%ymm0 , %%ymm9 \n\t" + "vmulpd %%ymm1, %%ymm14, %%ymm14 \n\t" + "vaddsubpd %%ymm14 , %%ymm0 , %%ymm10 \n\t" + "vmulpd %%ymm1, %%ymm15, %%ymm15 \n\t" + "vaddsubpd %%ymm15 , %%ymm0 , %%ymm11 \n\t" + + "vmovups %%ymm8 , -128(%1) \n\t" + "vmovups %%ymm9 , -96(%1) \n\t" + "vmovups %%ymm10, -64(%1) \n\t" + "vmovups %%ymm11, -32(%1) \n\t" + + "vzeroupper \n\t" + + : + : + "r" (n), // 0 + "r" (x), // 1 + "r" (alpha) // 2 + : "cc", "%0", "%1", + "%xmm0", "%xmm1", "%xmm2", "%xmm3", + "%xmm4", "%xmm5", "%xmm6", "%xmm7", + "%xmm8", "%xmm9", "%xmm10", "%xmm11", + "%xmm12", "%xmm13", "%xmm14", "%xmm15", + "memory" + ); + +} + + + +static void zscal_kernel_8_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline)); + +static void zscal_kernel_8_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x) +{ + + + __asm__ __volatile__ + ( + "vbroadcastsd (%2), %%ymm0 \n\t" // da_r + + "addq $128, %1 \n\t" + + "vmovups -128(%1), %%ymm4 \n\t" + "vmovups -96(%1), %%ymm5 \n\t" + "vmovups -64(%1), %%ymm6 \n\t" + "vmovups -32(%1), %%ymm7 \n\t" + + + "subq $8 , %0 \n\t" + "jz 2f \n\t" + + ".align 16 \n\t" + "1: \n\t" + + //"prefetcht0 128(%1) \n\t" + // ".align 2 \n\t" + + "vmulpd %%ymm0, %%ymm4 , %%ymm8 \n\t" // da_r*x0 , da_r *x1 + "vmovups 0(%1), %%ymm4 \n\t" + "vmulpd %%ymm0, %%ymm5 , %%ymm9 \n\t" + "vmovups 32(%1), %%ymm5 \n\t" + "vmulpd %%ymm0, %%ymm6 , %%ymm10 \n\t" + "vmovups 64(%1), %%ymm6 \n\t" + "vmulpd %%ymm0, %%ymm7 , %%ymm11 \n\t" + "vmovups 96(%1), %%ymm7 \n\t" + + "vmovups %%ymm8 , -128(%1) \n\t" + "vmovups %%ymm9 , -96(%1) \n\t" + "vmovups %%ymm10, -64(%1) \n\t" + "vmovups %%ymm11, -32(%1) \n\t" + + "addq $128 ,%1 \n\t" + "subq $8 , %0 \n\t" + "jnz 1b \n\t" + + "2: \n\t" + + + "vmulpd %%ymm0, %%ymm4 , %%ymm8 \n\t" // da_r*x0 , da_r *x1 + "vmulpd %%ymm0, %%ymm5 , %%ymm9 \n\t" + "vmulpd %%ymm0, %%ymm6 , %%ymm10 \n\t" + "vmulpd %%ymm0, %%ymm7 , %%ymm11 \n\t" + + "vmovups %%ymm8 , -128(%1) \n\t" + "vmovups %%ymm9 , -96(%1) \n\t" + "vmovups %%ymm10, -64(%1) \n\t" + "vmovups %%ymm11, -32(%1) \n\t" + + "vzeroupper \n\t" + + : + : + "r" (n), // 0 + "r" (x), // 1 + "r" (alpha) // 2 + : "cc", "%0", "%1", + "%xmm0", "%xmm1", "%xmm2", "%xmm3", + "%xmm4", "%xmm5", "%xmm6", "%xmm7", + "%xmm8", "%xmm9", "%xmm10", "%xmm11", + "%xmm12", "%xmm13", "%xmm14", "%xmm15", + "memory" + ); + +} + + +static void zscal_kernel_8_zero( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline)); + +static void zscal_kernel_8_zero( BLASLONG n, FLOAT *alpha, FLOAT *x) +{ + + + __asm__ __volatile__ + ( + "vxorpd %%ymm0, %%ymm0, %%ymm0 \n\t" + + "addq $128, %1 \n\t" + + ".align 16 \n\t" + "1: \n\t" + + //"prefetcht0 128(%1) \n\t" + // ".align 2 \n\t" + + "vmovups %%ymm0 , -128(%1) \n\t" + "vmovups %%ymm0 , -96(%1) \n\t" + "vmovups %%ymm0 , -64(%1) \n\t" + "vmovups %%ymm0 , -32(%1) \n\t" + + "addq $128 ,%1 \n\t" + "subq $8 , %0 \n\t" + "jnz 1b \n\t" + + "vzeroupper \n\t" + + : + : + "r" (n), // 0 + "r" (x), // 1 + "r" (alpha) // 2 + : "cc", "%0", "%1", + "%xmm0", "%xmm1", "%xmm2", "%xmm3", + "%xmm4", "%xmm5", "%xmm6", "%xmm7", + "%xmm8", "%xmm9", "%xmm10", "%xmm11", + "%xmm12", "%xmm13", "%xmm14", "%xmm15", + "memory" + ); + +} + + +