From 6eee1beac524b5582a6c6de14d9d35a78c1ece74 Mon Sep 17 00:00:00 2001
From: Andrew <16061801+brada4@users.noreply.github.com>
Date: Sun, 24 Feb 2019 20:41:02 +0200
Subject: [PATCH 2/2] move fix to right place
---
dgemv_n_microk_piledriver-4.c | 247 --------------------
kernel/x86_64/dgemv_n_microk_piledriver-4.c | 98 ++++----
2 files changed, 49 insertions(+), 296 deletions(-)
delete mode 100644 dgemv_n_microk_piledriver-4.c
diff --git a/dgemv_n_microk_piledriver-4.c b/dgemv_n_microk_piledriver-4.c
deleted file mode 100644
index 466931b82..000000000
--- a/dgemv_n_microk_piledriver-4.c
+++ /dev/null
@@ -1,247 +0,0 @@
-/***************************************************************************
-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_4x8 1
-static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLONG lda4, FLOAT *alpha) __attribute__ ((noinline));
-
-static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLONG lda4, FLOAT *alpha)
-{
-
- BLASLONG register i = 0;
-
- __asm__ __volatile__
- (
- "vzeroupper \n\t"
- "vbroadcastsd (%3), %%ymm12 \n\t" // x0
- "vbroadcastsd 8(%3), %%ymm13 \n\t" // x1
- "vbroadcastsd 16(%3), %%ymm14 \n\t" // x2
- "vbroadcastsd 24(%3), %%ymm15 \n\t" // x3
- "vbroadcastsd 32(%3), %%ymm0 \n\t" // x4
- "vbroadcastsd 40(%3), %%ymm1 \n\t" // x5
- "vbroadcastsd 48(%3), %%ymm2 \n\t" // x6
- "vbroadcastsd 56(%3), %%ymm3 \n\t" // x7
-
- "vbroadcastsd (%9), %%ymm6 \n\t" // alpha
-
- "testq $0x04, %1 \n\t"
- "jz 2f \n\t"
-
- "vmovupd (%4,%0,8), %%ymm7 \n\t" // 4 * y
- "vxorpd %%ymm4 , %%ymm4, %%ymm4 \n\t"
- "vxorpd %%ymm5 , %%ymm5, %%ymm5 \n\t"
-
- "vfmadd231pd (%5,%0,8), %%ymm12, %%ymm4 \n\t"
- "vfmadd231pd (%6,%0,8), %%ymm13, %%ymm5 \n\t"
- "vfmadd231pd (%7,%0,8), %%ymm14, %%ymm4 \n\t"
- "vfmadd231pd (%8,%0,8), %%ymm15, %%ymm5 \n\t"
-
- "vfmadd231pd (%5,%2,8), %%ymm0 , %%ymm4 \n\t"
- "vfmadd231pd (%6,%2,8), %%ymm1 , %%ymm5 \n\t"
- "vfmadd231pd (%7,%2,8), %%ymm2 , %%ymm4 \n\t"
- "vfmadd231pd (%8,%2,8), %%ymm3 , %%ymm5 \n\t"
-
- "vaddpd %%ymm4 , %%ymm5 , %%ymm5 \n\t"
- "vmulpd %%ymm6 , %%ymm5 , %%ymm5 \n\t"
- "vaddpd %%ymm7 , %%ymm5 , %%ymm5 \n\t"
-
-
- "vmovupd %%ymm5, (%4,%0,8) \n\t" // 4 * y
-
- "addq $4 , %2 \n\t"
- "addq $4 , %0 \n\t"
- "subq $4 , %1 \n\t"
-
- "2: \n\t"
-
- "cmpq $0, %1 \n\t"
- "je 3f \n\t"
-
-
- ".align 16 \n\t"
- "1: \n\t"
-
- "vxorpd %%ymm4 , %%ymm4, %%ymm4 \n\t"
- "vxorpd %%ymm5 , %%ymm5, %%ymm5 \n\t"
- "vmovupd (%4,%0,8), %%ymm8 \n\t" // 4 * y
- "vmovupd 32(%4,%0,8), %%ymm9 \n\t" // 4 * y
-
- "vfmadd231pd (%5,%0,8), %%ymm12, %%ymm4 \n\t"
- "vfmadd231pd 32(%5,%0,8), %%ymm12, %%ymm5 \n\t"
- "vfmadd231pd (%6,%0,8), %%ymm13, %%ymm4 \n\t"
- "vfmadd231pd 32(%6,%0,8), %%ymm13, %%ymm5 \n\t"
- "vfmadd231pd (%7,%0,8), %%ymm14, %%ymm4 \n\t"
- "vfmadd231pd 32(%7,%0,8), %%ymm14, %%ymm5 \n\t"
- "vfmadd231pd (%8,%0,8), %%ymm15, %%ymm4 \n\t"
- "vfmadd231pd 32(%8,%0,8), %%ymm15, %%ymm5 \n\t"
-
- "vfmadd231pd (%5,%2,8), %%ymm0 , %%ymm4 \n\t"
- "addq $8 , %0 \n\t"
- "vfmadd231pd 32(%5,%2,8), %%ymm0 , %%ymm5 \n\t"
- "vfmadd231pd (%6,%2,8), %%ymm1 , %%ymm4 \n\t"
- "vfmadd231pd 32(%6,%2,8), %%ymm1 , %%ymm5 \n\t"
- "vfmadd231pd (%7,%2,8), %%ymm2 , %%ymm4 \n\t"
- "vfmadd231pd 32(%7,%2,8), %%ymm2 , %%ymm5 \n\t"
- "vfmadd231pd (%8,%2,8), %%ymm3 , %%ymm4 \n\t"
- "vfmadd231pd 32(%8,%2,8), %%ymm3 , %%ymm5 \n\t"
-
- "vfmadd231pd %%ymm6 , %%ymm4 , %%ymm8 \n\t"
- "vfmadd231pd %%ymm6 , %%ymm5 , %%ymm9 \n\t"
-
- "addq $8 , %2 \n\t"
- "vmovupd %%ymm8,-64(%3,%0,8) \n\t" // 4 * y
- "subq $8 , %1 \n\t"
- "vmovupd %%ymm9,-32(%4,%0,8) \n\t" // 4 * y
-
- "jnz 1b \n\t"
-
- "3: \n\t"
- "vzeroupper \n\t"
-
- :
- "+r" (i), // 0
- "+r" (n), // 1
- "+r" (lda4) // 2
- :
- "r" (x), // 3
- "r" (y), // 4
- "r" (ap[0]), // 5
- "r" (ap[1]), // 6
- "r" (ap[2]), // 7
- "r" (ap[3]), // 8
- "r" (alpha) // 9
- : "cc",
- "%xmm0", "%xmm1",
- "%xmm2", "%xmm3",
- "%xmm4", "%xmm5",
- "%xmm6", "%xmm7",
- "%xmm8", "%xmm9",
- "%xmm12", "%xmm13", "%xmm14", "%xmm15",
- "memory"
- );
-
-}
-
-
-
-#define HAVE_KERNEL_4x4 1
-static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha) __attribute__ ((noinline));
-
-static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha)
-{
-
- BLASLONG register i = 0;
-
- __asm__ __volatile__
- (
- "vzeroupper \n\t"
- "vbroadcastsd (%2), %%ymm12 \n\t" // x0
- "vbroadcastsd 8(%2), %%ymm13 \n\t" // x1
- "vbroadcastsd 16(%2), %%ymm14 \n\t" // x2
- "vbroadcastsd 24(%2), %%ymm15 \n\t" // x3
-
- "vbroadcastsd (%8), %%ymm6 \n\t" // alpha
-
- "testq $0x04, %1 \n\t"
- "jz 2f \n\t"
-
- "vxorpd %%ymm4 , %%ymm4, %%ymm4 \n\t"
- "vxorpd %%ymm5 , %%ymm5, %%ymm5 \n\t"
- "vmovupd (%3,%0,8), %%ymm7 \n\t" // 4 * y
-
- "vfmadd231pd (%4,%0,8), %%ymm12, %%ymm4 \n\t"
- "vfmadd231pd (%5,%0,8), %%ymm13, %%ymm5 \n\t"
- "vfmadd231pd (%6,%0,8), %%ymm14, %%ymm4 \n\t"
- "vfmadd231pd (%7,%0,8), %%ymm15, %%ymm5 \n\t"
-
- "vaddpd %%ymm4 , %%ymm5 , %%ymm5 \n\t"
- "vmulpd %%ymm6 , %%ymm5 , %%ymm5 \n\t"
- "vaddpd %%ymm7 , %%ymm5 , %%ymm5 \n\t"
-
- "vmovupd %%ymm5, (%3,%0,8) \n\t" // 4 * y
-
- "addq $4 , %0 \n\t"
- "subq $4 , %1 \n\t"
-
- "2: \n\t"
-
- "cmpq $0, %1 \n\t"
- "je 3f \n\t"
-
-
- ".align 16 \n\t"
- "1: \n\t"
- "vxorpd %%ymm4 , %%ymm4, %%ymm4 \n\t"
- "vxorpd %%ymm5 , %%ymm5, %%ymm5 \n\t"
- "vmovupd (%3,%0,8), %%ymm8 \n\t" // 4 * y
- "vmovupd 32(%3,%0,8), %%ymm9 \n\t" // 4 * y
-
- "vfmadd231pd (%4,%0,8), %%ymm12, %%ymm4 \n\t"
- "vfmadd231pd 32(%4,%0,8), %%ymm12, %%ymm5 \n\t"
- "vfmadd231pd (%5,%0,8), %%ymm13, %%ymm4 \n\t"
- "vfmadd231pd 32(%5,%0,8), %%ymm13, %%ymm5 \n\t"
- "vfmadd231pd (%6,%0,8), %%ymm14, %%ymm4 \n\t"
- "vfmadd231pd 32(%6,%0,8), %%ymm14, %%ymm5 \n\t"
- "vfmadd231pd (%7,%0,8), %%ymm15, %%ymm4 \n\t"
- "vfmadd231pd 32(%7,%0,8), %%ymm15, %%ymm5 \n\t"
-
- "vfmadd231pd %%ymm6 , %%ymm4 , %%ymm8 \n\t"
- "vfmadd231pd %%ymm6 , %%ymm5 , %%ymm9 \n\t"
-
- "vmovupd %%ymm8, (%3,%0,8) \n\t" // 4 * y
- "vmovupd %%ymm9, 32(%3,%0,8) \n\t" // 4 * y
-
- "addq $8 , %0 \n\t"
- "subq $8 , %1 \n\t"
- "jnz 1b \n\t"
-
- "3: \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
- "r" (alpha) // 8
- : "cc",
- "%xmm4", "%xmm5",
- "%xmm6", "%xmm7",
- "%xmm8", "%xmm9",
- "%xmm12", "%xmm13", "%xmm14", "%xmm15",
- "memory"
- );
-
-}
-
-
diff --git a/kernel/x86_64/dgemv_n_microk_piledriver-4.c b/kernel/x86_64/dgemv_n_microk_piledriver-4.c
index 530780bab..466931b82 100644
--- a/kernel/x86_64/dgemv_n_microk_piledriver-4.c
+++ b/kernel/x86_64/dgemv_n_microk_piledriver-4.c
@@ -38,42 +38,42 @@ static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
__asm__ __volatile__
(
"vzeroupper \n\t"
- "vbroadcastsd (%2), %%ymm12 \n\t" // x0
- "vbroadcastsd 8(%2), %%ymm13 \n\t" // x1
- "vbroadcastsd 16(%2), %%ymm14 \n\t" // x2
- "vbroadcastsd 24(%2), %%ymm15 \n\t" // x3
- "vbroadcastsd 32(%2), %%ymm0 \n\t" // x4
- "vbroadcastsd 40(%2), %%ymm1 \n\t" // x5
- "vbroadcastsd 48(%2), %%ymm2 \n\t" // x6
- "vbroadcastsd 56(%2), %%ymm3 \n\t" // x7
+ "vbroadcastsd (%3), %%ymm12 \n\t" // x0
+ "vbroadcastsd 8(%3), %%ymm13 \n\t" // x1
+ "vbroadcastsd 16(%3), %%ymm14 \n\t" // x2
+ "vbroadcastsd 24(%3), %%ymm15 \n\t" // x3
+ "vbroadcastsd 32(%3), %%ymm0 \n\t" // x4
+ "vbroadcastsd 40(%3), %%ymm1 \n\t" // x5
+ "vbroadcastsd 48(%3), %%ymm2 \n\t" // x6
+ "vbroadcastsd 56(%3), %%ymm3 \n\t" // x7
"vbroadcastsd (%9), %%ymm6 \n\t" // alpha
"testq $0x04, %1 \n\t"
"jz 2f \n\t"
- "vmovupd (%3,%0,8), %%ymm7 \n\t" // 4 * y
+ "vmovupd (%4,%0,8), %%ymm7 \n\t" // 4 * y
"vxorpd %%ymm4 , %%ymm4, %%ymm4 \n\t"
"vxorpd %%ymm5 , %%ymm5, %%ymm5 \n\t"
- "vfmadd231pd (%4,%0,8), %%ymm12, %%ymm4 \n\t"
- "vfmadd231pd (%5,%0,8), %%ymm13, %%ymm5 \n\t"
- "vfmadd231pd (%6,%0,8), %%ymm14, %%ymm4 \n\t"
- "vfmadd231pd (%7,%0,8), %%ymm15, %%ymm5 \n\t"
+ "vfmadd231pd (%5,%0,8), %%ymm12, %%ymm4 \n\t"
+ "vfmadd231pd (%6,%0,8), %%ymm13, %%ymm5 \n\t"
+ "vfmadd231pd (%7,%0,8), %%ymm14, %%ymm4 \n\t"
+ "vfmadd231pd (%8,%0,8), %%ymm15, %%ymm5 \n\t"
- "vfmadd231pd (%4,%8,8), %%ymm0 , %%ymm4 \n\t"
- "vfmadd231pd (%5,%8,8), %%ymm1 , %%ymm5 \n\t"
- "vfmadd231pd (%6,%8,8), %%ymm2 , %%ymm4 \n\t"
- "vfmadd231pd (%7,%8,8), %%ymm3 , %%ymm5 \n\t"
+ "vfmadd231pd (%5,%2,8), %%ymm0 , %%ymm4 \n\t"
+ "vfmadd231pd (%6,%2,8), %%ymm1 , %%ymm5 \n\t"
+ "vfmadd231pd (%7,%2,8), %%ymm2 , %%ymm4 \n\t"
+ "vfmadd231pd (%8,%2,8), %%ymm3 , %%ymm5 \n\t"
"vaddpd %%ymm4 , %%ymm5 , %%ymm5 \n\t"
"vmulpd %%ymm6 , %%ymm5 , %%ymm5 \n\t"
"vaddpd %%ymm7 , %%ymm5 , %%ymm5 \n\t"
- "vmovupd %%ymm5, (%3,%0,8) \n\t" // 4 * y
+ "vmovupd %%ymm5, (%4,%0,8) \n\t" // 4 * y
- "addq $4 , %8 \n\t"
+ "addq $4 , %2 \n\t"
"addq $4 , %0 \n\t"
"subq $4 , %1 \n\t"
@@ -88,35 +88,35 @@ static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
"vxorpd %%ymm4 , %%ymm4, %%ymm4 \n\t"
"vxorpd %%ymm5 , %%ymm5, %%ymm5 \n\t"
- "vmovupd (%3,%0,8), %%ymm8 \n\t" // 4 * y
- "vmovupd 32(%3,%0,8), %%ymm9 \n\t" // 4 * y
-
- "vfmadd231pd (%4,%0,8), %%ymm12, %%ymm4 \n\t"
- "vfmadd231pd 32(%4,%0,8), %%ymm12, %%ymm5 \n\t"
- "vfmadd231pd (%5,%0,8), %%ymm13, %%ymm4 \n\t"
- "vfmadd231pd 32(%5,%0,8), %%ymm13, %%ymm5 \n\t"
- "vfmadd231pd (%6,%0,8), %%ymm14, %%ymm4 \n\t"
- "vfmadd231pd 32(%6,%0,8), %%ymm14, %%ymm5 \n\t"
- "vfmadd231pd (%7,%0,8), %%ymm15, %%ymm4 \n\t"
- "vfmadd231pd 32(%7,%0,8), %%ymm15, %%ymm5 \n\t"
-
- "vfmadd231pd (%4,%8,8), %%ymm0 , %%ymm4 \n\t"
+ "vmovupd (%4,%0,8), %%ymm8 \n\t" // 4 * y
+ "vmovupd 32(%4,%0,8), %%ymm9 \n\t" // 4 * y
+
+ "vfmadd231pd (%5,%0,8), %%ymm12, %%ymm4 \n\t"
+ "vfmadd231pd 32(%5,%0,8), %%ymm12, %%ymm5 \n\t"
+ "vfmadd231pd (%6,%0,8), %%ymm13, %%ymm4 \n\t"
+ "vfmadd231pd 32(%6,%0,8), %%ymm13, %%ymm5 \n\t"
+ "vfmadd231pd (%7,%0,8), %%ymm14, %%ymm4 \n\t"
+ "vfmadd231pd 32(%7,%0,8), %%ymm14, %%ymm5 \n\t"
+ "vfmadd231pd (%8,%0,8), %%ymm15, %%ymm4 \n\t"
+ "vfmadd231pd 32(%8,%0,8), %%ymm15, %%ymm5 \n\t"
+
+ "vfmadd231pd (%5,%2,8), %%ymm0 , %%ymm4 \n\t"
"addq $8 , %0 \n\t"
- "vfmadd231pd 32(%4,%8,8), %%ymm0 , %%ymm5 \n\t"
- "vfmadd231pd (%5,%8,8), %%ymm1 , %%ymm4 \n\t"
- "vfmadd231pd 32(%5,%8,8), %%ymm1 , %%ymm5 \n\t"
- "vfmadd231pd (%6,%8,8), %%ymm2 , %%ymm4 \n\t"
- "vfmadd231pd 32(%6,%8,8), %%ymm2 , %%ymm5 \n\t"
- "vfmadd231pd (%7,%8,8), %%ymm3 , %%ymm4 \n\t"
- "vfmadd231pd 32(%7,%8,8), %%ymm3 , %%ymm5 \n\t"
+ "vfmadd231pd 32(%5,%2,8), %%ymm0 , %%ymm5 \n\t"
+ "vfmadd231pd (%6,%2,8), %%ymm1 , %%ymm4 \n\t"
+ "vfmadd231pd 32(%6,%2,8), %%ymm1 , %%ymm5 \n\t"
+ "vfmadd231pd (%7,%2,8), %%ymm2 , %%ymm4 \n\t"
+ "vfmadd231pd 32(%7,%2,8), %%ymm2 , %%ymm5 \n\t"
+ "vfmadd231pd (%8,%2,8), %%ymm3 , %%ymm4 \n\t"
+ "vfmadd231pd 32(%8,%2,8), %%ymm3 , %%ymm5 \n\t"
"vfmadd231pd %%ymm6 , %%ymm4 , %%ymm8 \n\t"
"vfmadd231pd %%ymm6 , %%ymm5 , %%ymm9 \n\t"
- "addq $8 , %8 \n\t"
+ "addq $8 , %2 \n\t"
"vmovupd %%ymm8,-64(%3,%0,8) \n\t" // 4 * y
"subq $8 , %1 \n\t"
- "vmovupd %%ymm9,-32(%3,%0,8) \n\t" // 4 * y
+ "vmovupd %%ymm9,-32(%4,%0,8) \n\t" // 4 * y
"jnz 1b \n\t"
@@ -125,15 +125,15 @@ static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
:
"+r" (i), // 0
- "+r" (n) // 1
+ "+r" (n), // 1
+ "+r" (lda4) // 2
:
- "r" (x), // 2
- "r" (y), // 3
- "r" (ap[0]), // 4
- "r" (ap[1]), // 5
- "r" (ap[2]), // 6
- "r" (ap[3]), // 7
- "r" (lda4), // 8
+ "r" (x), // 3
+ "r" (y), // 4
+ "r" (ap[0]), // 5
+ "r" (ap[1]), // 6
+ "r" (ap[2]), // 7
+ "r" (ap[3]), // 8
"r" (alpha) // 9
: "cc",
"%xmm0", "%xmm1",