diff --git a/CONTRIBUTORS.md b/CONTRIBUTORS.md
index ebe52ea8a..4431103bd 100644
--- a/CONTRIBUTORS.md
+++ b/CONTRIBUTORS.md
@@ -150,3 +150,6 @@ In chronological order:
* theoractice
* [2016-03-20] Fix compiler error in VisualStudio with CMake
* [2016-03-22] Fix access violation on Windows while static linking
+
+* Shivraj Patil
+ * [2016-05-03] DGEMM optimization for MIPS P5600 and I6400 using MSA
diff --git a/kernel/mips/KERNEL.P5600 b/kernel/mips/KERNEL.P5600
index 09064fe48..d21575251 100644
--- a/kernel/mips/KERNEL.P5600
+++ b/kernel/mips/KERNEL.P5600
@@ -91,9 +91,13 @@ SGEMMOTCOPY = ../generic/gemm_tcopy_2.c
SGEMMONCOPYOBJ = sgemm_oncopy.o
SGEMMOTCOPYOBJ = sgemm_otcopy.o
-DGEMMKERNEL = ../generic/gemmkernel_2x2.c
-DGEMMONCOPY = ../generic/gemm_ncopy_2.c
-DGEMMOTCOPY = ../generic/gemm_tcopy_2.c
+DGEMMKERNEL = ../mips/dgemm_kernel_8x4_msa.c
+DGEMMINCOPY = ../mips/dgemm_ncopy_8_msa.c
+DGEMMITCOPY = ../mips/dgemm_tcopy_8_msa.c
+DGEMMONCOPY = ../mips/dgemm_ncopy_4_msa.c
+DGEMMOTCOPY = ../mips/dgemm_tcopy_4_msa.c
+DGEMMINCOPYOBJ = dgemm_incopy.o
+DGEMMITCOPYOBJ = dgemm_itcopy.o
DGEMMONCOPYOBJ = dgemm_oncopy.o
DGEMMOTCOPYOBJ = dgemm_otcopy.o
diff --git a/kernel/mips/dgemm_kernel_8x4_msa.c b/kernel/mips/dgemm_kernel_8x4_msa.c
new file mode 100644
index 000000000..8d9e3455e
--- /dev/null
+++ b/kernel/mips/dgemm_kernel_8x4_msa.c
@@ -0,0 +1,720 @@
+/*******************************************************************************
+Copyright (c) 2016, 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.
+*******************************************************************************/
+
+#include "common.h"
+#include "macros_msa.h"
+
+int CNAME(BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT *A, FLOAT *B,
+ FLOAT *C, BLASLONG ldc
+#ifdef TRMMKERNEL
+ , BLASLONG offset
+#endif
+ )
+{
+ BLASLONG i, j, l;
+ FLOAT *pc0, *pc1, *pc2, *pc3;
+ FLOAT *pa0, *pb0;
+ FLOAT tmp0, tmp1, tmp2, tmp3;
+ FLOAT a0;
+ FLOAT b0, b1, b2, b3;
+ v2f64 v_alpha = {alpha, alpha};
+ v2f64 src_a0, src_a1, src_a2, src_a3, src_b, src_b0, src_b1;
+ v2f64 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+ v2f64 res0, res1, res2, res3, res4, res5, res6, res7;
+ v2f64 res8, res9, res10, res11, res12, res13, res14, res15;
+
+ for (j = (n / 4); j--;)
+ {
+ pc0 = C;
+ pc1 = pc0 + ldc;
+ pc2 = pc1 + ldc;
+ pc3 = pc2 + ldc;
+
+ pa0 = A;
+
+ for (i = (m / 8); i--;)
+ {
+ pb0 = B;
+
+ LD_DP4(pa0, 2, src_a0, src_a1, src_a2, src_a3);
+ LD_DP2(pb0, 2, src_b0, src_b1);
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b0, (v2i64) src_b0);
+ res0 = src_a0 * src_b;
+ res1 = src_a1 * src_b;
+ res2 = src_a2 * src_b;
+ res3 = src_a3 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b0, (v2i64) src_b0);
+ res4 = src_a0 * src_b;
+ res5 = src_a1 * src_b;
+ res6 = src_a2 * src_b;
+ res7 = src_a3 * src_b;
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b1, (v2i64) src_b1);
+ res8 = src_a0 * src_b;
+ res9 = src_a1 * src_b;
+ res10 = src_a2 * src_b;
+ res11 = src_a3 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b1, (v2i64) src_b1);
+ res12 = src_a0 * src_b;
+ res13 = src_a1 * src_b;
+ res14 = src_a2 * src_b;
+ res15 = src_a3 * src_b;
+
+ pa0 += 8;
+ pb0 += 4;
+
+ for (l = (k - 1); l--;)
+ {
+ LD_DP4(pa0, 2, src_a0, src_a1, src_a2, src_a3);
+ LD_DP2(pb0, 2, src_b0, src_b1);
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b0, (v2i64) src_b0);
+ res0 += src_a0 * src_b;
+ res1 += src_a1 * src_b;
+ res2 += src_a2 * src_b;
+ res3 += src_a3 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b0, (v2i64) src_b0);
+ res4 += src_a0 * src_b;
+ res5 += src_a1 * src_b;
+ res6 += src_a2 * src_b;
+ res7 += src_a3 * src_b;
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b1, (v2i64) src_b1);
+ res8 += src_a0 * src_b;
+ res9 += src_a1 * src_b;
+ res10 += src_a2 * src_b;
+ res11 += src_a3 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b1, (v2i64) src_b1);
+ res12 += src_a0 * src_b;
+ res13 += src_a1 * src_b;
+ res14 += src_a2 * src_b;
+ res15 += src_a3 * src_b;
+
+ pa0 += 8;
+ pb0 += 4;
+ }
+
+ LD_DP4(pc0, 2, dst0, dst1, dst2, dst3);
+ LD_DP4(pc1, 2, dst4, dst5, dst6, dst7);
+
+ dst0 += res0 * v_alpha;
+ dst1 += res1 * v_alpha;
+ dst2 += res2 * v_alpha;
+ dst3 += res3 * v_alpha;
+ dst4 += res4 * v_alpha;
+ dst5 += res5 * v_alpha;
+ dst6 += res6 * v_alpha;
+ dst7 += res7 * v_alpha;
+
+ ST_DP4(dst0, dst1, dst2, dst3, pc0, 2);
+ ST_DP4(dst4, dst5, dst6, dst7, pc1, 2);
+
+ LD_DP4(pc2, 2, dst0, dst1, dst2, dst3);
+ LD_DP4(pc3, 2, dst4, dst5, dst6, dst7);
+
+ dst0 += res8 * v_alpha;
+ dst1 += res9 * v_alpha;
+ dst2 += res10 * v_alpha;
+ dst3 += res11 * v_alpha;
+ dst4 += res12 * v_alpha;
+ dst5 += res13 * v_alpha;
+ dst6 += res14 * v_alpha;
+ dst7 += res15 * v_alpha;
+
+ ST_DP4(dst0, dst1, dst2, dst3, pc2, 2);
+ ST_DP4(dst4, dst5, dst6, dst7, pc3, 2);
+
+ pc0 += 8;
+ pc1 += 8;
+ pc2 += 8;
+ pc3 += 8;
+ }
+
+ for (i = ((m & 4) / 4); i--;)
+ {
+ pb0 = B;
+
+ LD_DP2(pa0, 2, src_a0, src_a1);
+ LD_DP2(pb0, 2, src_b0, src_b1);
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b0, (v2i64) src_b0);
+ res0 = src_a0 * src_b;
+ res1 = src_a1 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b0, (v2i64) src_b0);
+ res2 = src_a0 * src_b;
+ res3 = src_a1 * src_b;
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b1, (v2i64) src_b1);
+ res4 = src_a0 * src_b;
+ res5 = src_a1 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b1, (v2i64) src_b1);
+ res6 = src_a0 * src_b;
+ res7 = src_a1 * src_b;
+
+ pa0 += 4;
+ pb0 += 4;
+
+ for (l = (k - 1); l--;)
+ {
+ LD_DP2(pa0, 2, src_a0, src_a1);
+ LD_DP2(pb0, 2, src_b0, src_b1);
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b0, (v2i64) src_b0);
+ res0 += src_a0 * src_b;
+ res1 += src_a1 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b0, (v2i64) src_b0);
+ res2 += src_a0 * src_b;
+ res3 += src_a1 * src_b;
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b1, (v2i64) src_b1);
+ res4 += src_a0 * src_b;
+ res5 += src_a1 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b1, (v2i64) src_b1);
+ res6 += src_a0 * src_b;
+ res7 += src_a1 * src_b;
+
+ pa0 += 4;
+ pb0 += 4;
+ }
+
+ LD_DP2(pc0, 2, dst0, dst1);
+ LD_DP2(pc1, 2, dst2, dst3);
+ LD_DP2(pc2, 2, dst4, dst5);
+ LD_DP2(pc3, 2, dst6, dst7);
+
+ dst0 += res0 * v_alpha;
+ dst1 += res1 * v_alpha;
+ dst2 += res2 * v_alpha;
+ dst3 += res3 * v_alpha;
+ dst4 += res4 * v_alpha;
+ dst5 += res5 * v_alpha;
+ dst6 += res6 * v_alpha;
+ dst7 += res7 * v_alpha;
+
+ ST_DP2(dst0, dst1, pc0, 2);
+ ST_DP2(dst2, dst3, pc1, 2);
+ ST_DP2(dst4, dst5, pc2, 2);
+ ST_DP2(dst6, dst7, pc3, 2);
+
+ pc0 += 4;
+ pc1 += 4;
+ pc2 += 4;
+ pc3 += 4;
+ }
+
+ for (i = ((m & 2) / 2); i--;)
+ {
+ pb0 = B;
+
+ src_a0 = LD_DP(pa0);
+ LD_DP2(pb0, 2, src_b0, src_b1);
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b0, (v2i64) src_b0);
+ res0 = src_a0 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b0, (v2i64) src_b0);
+ res1 = src_a0 * src_b;
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b1, (v2i64) src_b1);
+ res2 = src_a0 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b1, (v2i64) src_b1);
+ res3 = src_a0 * src_b;
+
+ pa0 += 2;
+ pb0 += 4;
+
+ for (l = (k - 1); l--;)
+ {
+ src_a0 = LD_DP(pa0);
+ LD_DP2(pb0, 2, src_b0, src_b1);
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b0, (v2i64) src_b0);
+ res0 += src_a0 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b0, (v2i64) src_b0);
+ res1 += src_a0 * src_b;
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b1, (v2i64) src_b1);
+ res2 += src_a0 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b1, (v2i64) src_b1);
+ res3 += src_a0 * src_b;
+
+ pa0 += 2;
+ pb0 += 4;
+ }
+
+ dst0 = LD_DP(pc0);
+ dst1 = LD_DP(pc1);
+ dst2 = LD_DP(pc2);
+ dst3 = LD_DP(pc3);
+
+ dst0 += res0 * v_alpha;
+ dst1 += res1 * v_alpha;
+ dst2 += res2 * v_alpha;
+ dst3 += res3 * v_alpha;
+
+ ST_DP(dst0, pc0);
+ ST_DP(dst1, pc1);
+ ST_DP(dst2, pc2);
+ ST_DP(dst3, pc3);
+
+ pc0 += 2;
+ pc1 += 2;
+ pc2 += 2;
+ pc3 += 2;
+ }
+
+ for (i = (m & 1); i--;)
+ {
+ pb0 = B;
+
+ a0 = pa0[0];
+ b0 = pb0[0];
+ tmp0 = a0 * b0;
+
+ b1 = pb0[1];
+ tmp1 = a0 * b1;
+
+ b2 = pb0[2];
+ tmp2 = a0 * b2;
+
+ b3 = pb0[3];
+ tmp3 = a0 * b3;
+
+ pa0 += 1;
+ pb0 += 4;
+
+ for (l = (k - 1); l--;)
+ {
+ a0 = pa0[0];
+ b0 = pb0[0];
+ tmp0 += a0 * b0;
+
+ b1 = pb0[1];
+ tmp1 += a0 * b1;
+
+ b2 = pb0[2];
+ tmp2 += a0 * b2;
+
+ b3 = pb0[3];
+ tmp3 += a0 * b3;
+
+ pa0 += 1;
+ pb0 += 4;
+ }
+
+ tmp0 = alpha * tmp0;
+ tmp1 = alpha * tmp1;
+ tmp2 = alpha * tmp2;
+ tmp3 = alpha * tmp3;
+
+ pc0[0] += tmp0;
+ pc1[0] += tmp1;
+ pc2[0] += tmp2;
+ pc3[0] += tmp3;
+
+ pc0 += 1;
+ pc1 += 1;
+ pc2 += 1;
+ pc3 += 1;
+ }
+
+ l = (k << 2);
+ B = B + l;
+ i = (ldc << 2);
+ C = C + i;
+ }
+
+ for (j = ((n & 2) / 2); j--;)
+ {
+ pc0 = C;
+ pc1 = pc0 + ldc;
+
+ pa0 = A;
+
+ for (i = (m / 8); i--;)
+ {
+ pb0 = B;
+
+ LD_DP4(pa0, 2, src_a0, src_a1, src_a2, src_a3);
+ src_b0 = LD_DP(pb0);
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b0, (v2i64) src_b0);
+ res0 = src_a0 * src_b;
+ res1 = src_a1 * src_b;
+ res2 = src_a2 * src_b;
+ res3 = src_a3 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b0, (v2i64) src_b0);
+ res4 = src_a0 * src_b;
+ res5 = src_a1 * src_b;
+ res6 = src_a2 * src_b;
+ res7 = src_a3 * src_b;
+
+ pa0 += 8;
+ pb0 += 2;
+
+ for (l = (k - 1); l--;)
+ {
+ LD_DP4(pa0, 2, src_a0, src_a1, src_a2, src_a3);
+ src_b0 = LD_DP(pb0);
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b0, (v2i64) src_b0);
+ res0 += src_a0 * src_b;
+ res1 += src_a1 * src_b;
+ res2 += src_a2 * src_b;
+ res3 += src_a3 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b0, (v2i64) src_b0);
+ res4 += src_a0 * src_b;
+ res5 += src_a1 * src_b;
+ res6 += src_a2 * src_b;
+ res7 += src_a3 * src_b;
+
+ pa0 += 8;
+ pb0 += 2;
+ }
+
+ LD_DP4(pc0, 2, dst0, dst1, dst2, dst3);
+ LD_DP4(pc1, 2, dst4, dst5, dst6, dst7);
+
+ dst0 += res0 * v_alpha;
+ dst1 += res1 * v_alpha;
+ dst2 += res2 * v_alpha;
+ dst3 += res3 * v_alpha;
+ dst4 += res4 * v_alpha;
+ dst5 += res5 * v_alpha;
+ dst6 += res6 * v_alpha;
+ dst7 += res7 * v_alpha;
+
+ ST_DP4(dst0, dst1, dst2, dst3, pc0, 2);
+ ST_DP4(dst4, dst5, dst6, dst7, pc1, 2);
+
+ pc0 += 8;
+ pc1 += 8;
+ }
+
+ for (i = ((m & 4) / 4); i--;)
+ {
+ pb0 = B;
+
+ LD_DP2(pa0, 2, src_a0, src_a1);
+ src_b0 = LD_DP(pb0);
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b0, (v2i64) src_b0);
+ res0 = src_a0 * src_b;
+ res1 = src_a1 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b0, (v2i64) src_b0);
+ res2 = src_a0 * src_b;
+ res3 = src_a1 * src_b;
+
+ pa0 += 4;
+ pb0 += 2;
+
+ for (l = (k - 1); l--;)
+ {
+ LD_DP2(pa0, 2, src_a0, src_a1);
+ src_b0 = LD_DP(pb0);
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b0, (v2i64) src_b0);
+ res0 += src_a0 * src_b;
+ res1 += src_a1 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b0, (v2i64) src_b0);
+ res2 += src_a0 * src_b;
+ res3 += src_a1 * src_b;
+
+ pa0 += 4;
+ pb0 += 2;
+ }
+
+ LD_DP2(pc0, 2, dst0, dst1);
+ LD_DP2(pc1, 2, dst2, dst3);
+
+ dst0 += res0 * v_alpha;
+ dst1 += res1 * v_alpha;
+ dst2 += res2 * v_alpha;
+ dst3 += res3 * v_alpha;
+
+ ST_DP2(dst0, dst1, pc0, 2);
+ ST_DP2(dst2, dst3, pc1, 2);
+
+ pc0 += 4;
+ pc1 += 4;
+ }
+
+ for (i = ((m & 2) / 2); i--;)
+ {
+ pb0 = B;
+
+ src_a0 = LD_DP(pa0);
+ src_b0 = LD_DP(pb0);
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b0, (v2i64) src_b0);
+ res0 = src_a0 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b0, (v2i64) src_b0);
+ res1 = src_a0 * src_b;
+
+ pa0 += 2;
+ pb0 += 2;
+
+ for (l = (k - 1); l--;)
+ {
+ src_a0 = LD_DP(pa0);
+ src_b0 = LD_DP(pb0);
+
+ src_b = (v2f64) __msa_ilvr_d((v2i64) src_b0, (v2i64) src_b0);
+ res0 += src_a0 * src_b;
+
+ src_b = (v2f64) __msa_ilvl_d((v2i64) src_b0, (v2i64) src_b0);
+ res1 += src_a0 * src_b;
+
+ pa0 += 2;
+ pb0 += 2;
+ }
+
+ dst0 = LD_DP(pc0);
+ dst1 = LD_DP(pc1);
+
+ dst0 += res0 * v_alpha;
+ dst1 += res1 * v_alpha;
+
+ ST_DP(dst0, pc0);
+ ST_DP(dst1, pc1);
+
+ pc0 += 2;
+ pc1 += 2;
+ }
+
+ for (i = (m & 1); i--;)
+ {
+ pb0 = B;
+
+ a0 = pa0[0];
+ b0 = pb0[0];
+ tmp0 = a0 * b0;
+
+ b1 = pb0[1];
+ tmp1 = a0 * b1;
+
+ pa0 += 1;
+ pb0 += 2;
+
+ for (l = (k - 1); l--;)
+ {
+ a0 = pa0[0];
+ b0 = pb0[0];
+ tmp0 += a0 * b0;
+
+ b1 = pb0[1];
+ tmp1 += a0 * b1;
+
+ pa0 += 1;
+ pb0 += 2;
+ }
+
+ tmp0 = alpha * tmp0;
+ tmp1 = alpha * tmp1;
+
+ pc0[0] += tmp0;
+ pc1[0] += tmp1;
+
+ pc0 += 1;
+ pc1 += 1;
+ }
+
+ l = (k << 1);
+ B = B + l;
+ i = (ldc << 1);
+ C = C + i;
+ }
+
+ for (j = (n & 1); j--;)
+ {
+ pc0 = C;
+ pa0 = A;
+
+ for (i = (m / 8); i--;)
+ {
+ pb0 = B;
+
+ LD_DP4(pa0, 2, src_a0, src_a1, src_a2, src_a3);
+ src_b[0] = pb0[0];
+ src_b[1] = pb0[0];
+
+ res0 = src_a0 * src_b;
+ res1 = src_a1 * src_b;
+ res2 = src_a2 * src_b;
+ res3 = src_a3 * src_b;
+
+ pa0 += 8;
+ pb0 += 1;
+
+ for (l = (k - 1); l--;)
+ {
+ LD_DP4(pa0, 2, src_a0, src_a1, src_a2, src_a3);
+ src_b[0] = pb0[0];
+ src_b[1] = pb0[0];
+
+ res0 += src_a0 * src_b;
+ res1 += src_a1 * src_b;
+ res2 += src_a2 * src_b;
+ res3 += src_a3 * src_b;
+
+ pa0 += 8;
+ pb0 += 1;
+ }
+
+ LD_DP4(pc0, 2, dst0, dst1, dst2, dst3);
+
+ dst0 += res0 * v_alpha;
+ dst1 += res1 * v_alpha;
+ dst2 += res2 * v_alpha;
+ dst3 += res3 * v_alpha;
+
+ ST_DP4(dst0, dst1, dst2, dst3, pc0, 2);
+
+ pc0 += 8;
+ }
+
+ for (i = ((m & 4) / 4); i--;)
+ {
+ pb0 = B;
+
+ LD_DP2(pa0, 2, src_a0, src_a1);
+ src_b[0] = pb0[0];
+ src_b[1] = pb0[0];
+
+ res0 = src_a0 * src_b;
+ res1 = src_a1 * src_b;
+
+ pa0 += 4;
+ pb0 += 1;
+
+ for (l = (k - 1); l--;)
+ {
+ LD_DP2(pa0, 2, src_a0, src_a1);
+ src_b[0] = pb0[0];
+ src_b[1] = pb0[0];
+
+ res0 += src_a0 * src_b;
+ res1 += src_a1 * src_b;
+
+ pa0 += 4;
+ pb0 += 1;
+ }
+
+ LD_DP2(pc0, 2, dst0, dst1);
+
+ dst0 += res0 * v_alpha;
+ dst1 += res1 * v_alpha;
+
+ ST_DP2(dst0, dst1, pc0, 2);
+
+ pc0 += 4;
+ }
+
+ for (i = ((m & 2) / 2); i--;)
+ {
+ pb0 = B;
+
+ src_a0 = LD_DP(pa0);
+ src_b[0] = pb0[0];
+ src_b[1] = pb0[0];
+
+ res0 = src_a0 * src_b;
+
+ pa0 += 2;
+ pb0 += 1;
+
+ for (l = (k - 1); l--;)
+ {
+ src_a0 = LD_DP(pa0);
+ src_b[0] = pb0[0];
+ src_b[1] = pb0[0];
+
+ res0 += src_a0 * src_b;
+
+ pa0 += 2;
+ pb0 += 1;
+ }
+
+ dst0 = LD_DP(pc0);
+
+ dst0 += res0 * v_alpha;
+
+ ST_DP(dst0, pc0);
+
+ pc0 += 2;
+ }
+
+ for (i = (m & 1); i--;)
+ {
+ pb0 = B;
+
+ a0 = pa0[0];
+ b0 = pb0[0];
+ tmp0 = a0 * b0;
+
+ pa0 += 1;
+ pb0 += 1;
+
+ for (l = (k - 1); l--;)
+ {
+ a0 = pa0[0];
+ b0 = pb0[0];
+ tmp0 += a0 * b0;
+
+ pa0 += 1;
+ pb0 += 1;
+ }
+
+ pc0[0] += alpha * tmp0;
+
+ pc0 += 1;
+ }
+
+ l = (k << 0);
+ B = B + l;
+ i = (ldc << 0);
+ C = C + i;
+ }
+ return 0;
+}
diff --git a/kernel/mips/dgemm_ncopy_4_msa.c b/kernel/mips/dgemm_ncopy_4_msa.c
new file mode 100644
index 000000000..bbd76070f
--- /dev/null
+++ b/kernel/mips/dgemm_ncopy_4_msa.c
@@ -0,0 +1,135 @@
+/*******************************************************************************
+Copyright (c) 2016, 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.
+*******************************************************************************/
+
+#include "common.h"
+#include "macros_msa.h"
+
+int CNAME(BLASLONG m, BLASLONG n, FLOAT * __restrict src, BLASLONG lda,
+ FLOAT * __restrict dst)
+{
+ BLASLONG i, j;
+ FLOAT *psrc0, *psrc1, *psrc2, *psrc3, *psrc4;
+ FLOAT *pdst;
+ v2f64 src0, src1, src2, src3, src4, src5, src6, src7;
+ v2f64 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+
+ psrc0 = src;
+ pdst = dst;
+
+ for (j = (n >> 2); j--;)
+ {
+ psrc1 = psrc0;
+ psrc2 = psrc1 + lda;
+ psrc3 = psrc2 + lda;
+ psrc4 = psrc3 + lda;
+ psrc0 += 4 * lda;
+
+ for (i = (m >> 2); i--;)
+ {
+ LD_DP2(psrc1, 2, src0, src1);
+ LD_DP2(psrc2, 2, src2, src3);
+ LD_DP2(psrc3, 2, src4, src5);
+ LD_DP2(psrc4, 2, src6, src7);
+
+ psrc1 += 4;
+ psrc2 += 4;
+ psrc3 += 4;
+ psrc4 += 4;
+
+ dst0 = (v2f64) __msa_ilvr_d((v2i64) src2, (v2i64) src0);
+ dst1 = (v2f64) __msa_ilvr_d((v2i64) src6, (v2i64) src4);
+ dst2 = (v2f64) __msa_ilvr_d((v2i64) src3, (v2i64) src1);
+ dst3 = (v2f64) __msa_ilvr_d((v2i64) src7, (v2i64) src5);
+
+ dst4 = (v2f64) __msa_ilvl_d((v2i64) src2, (v2i64) src0);
+ dst5 = (v2f64) __msa_ilvl_d((v2i64) src6, (v2i64) src4);
+ dst6 = (v2f64) __msa_ilvl_d((v2i64) src3, (v2i64) src1);
+ dst7 = (v2f64) __msa_ilvl_d((v2i64) src7, (v2i64) src5);
+
+ ST_DP8(dst0, dst1, dst4, dst5, dst2, dst3, dst6, dst7, pdst, 2);
+ pdst += 16;
+ }
+
+ for (i = (m & 3); i--;)
+ {
+ *pdst++ = *psrc1++;
+ *pdst++ = *psrc2++;
+ *pdst++ = *psrc3++;
+ *pdst++ = *psrc4++;
+ }
+ }
+
+ if (n & 2)
+ {
+ psrc1 = psrc0;
+ psrc2 = psrc1 + lda;
+ psrc0 += 2 * lda;
+
+ for (i = (m >> 2); i--;)
+ {
+ LD_DP2(psrc1, 2, src0, src1);
+ LD_DP2(psrc2, 2, src2, src3);
+ psrc1 += 4;
+ psrc2 += 4;
+
+ dst0 = (v2f64) __msa_ilvr_d((v2i64) src2, (v2i64) src0);
+ dst1 = (v2f64) __msa_ilvr_d((v2i64) src3, (v2i64) src1);
+ dst4 = (v2f64) __msa_ilvl_d((v2i64) src2, (v2i64) src0);
+ dst5 = (v2f64) __msa_ilvl_d((v2i64) src3, (v2i64) src1);
+
+ ST_DP4(dst0, dst4, dst1, dst5, pdst, 2);
+ pdst += 8;
+ }
+
+ for (i = (m & 3); i--;)
+ {
+ *pdst++ = *psrc1++;
+ *pdst++ = *psrc2++;
+ }
+ }
+
+ if (n & 1)
+ {
+ psrc1 = psrc0;
+
+ for (i = (m >> 2); i--;)
+ {
+ LD_DP2(psrc1, 2, src0, src1);
+ psrc1 += 4;
+
+ ST_DP2(src0, src1, pdst, 2);
+ pdst += 4;
+ }
+
+ for (i = (m & 3); i--;)
+ {
+ *pdst++ = *psrc1++;
+ }
+ }
+
+ return 0;
+}
diff --git a/kernel/mips/dgemm_ncopy_8_msa.c b/kernel/mips/dgemm_ncopy_8_msa.c
new file mode 100644
index 000000000..43c977582
--- /dev/null
+++ b/kernel/mips/dgemm_ncopy_8_msa.c
@@ -0,0 +1,228 @@
+/*******************************************************************************
+Copyright (c) 2016, 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.
+*******************************************************************************/
+
+#include "common.h"
+#include "macros_msa.h"
+
+int CNAME(BLASLONG m, BLASLONG n, FLOAT * __restrict src, BLASLONG lda,
+ FLOAT * __restrict dst)
+{
+ BLASLONG i, j;
+ FLOAT *psrc0, *psrc1, *psrc2, *psrc3, *psrc4;
+ FLOAT *psrc5, *psrc6, *psrc7, *psrc8;
+ FLOAT *pdst;
+ v2f64 src0, src1, src2, src3, src4, src5, src6, src7;
+ v2f64 src8, src9, src10, src11, src12, src13, src14, src15;
+ v2f64 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+
+ psrc0 = src;
+ pdst = dst;
+
+ for (j = (n >> 3); j--;)
+ {
+ psrc1 = psrc0;
+ psrc2 = psrc1 + lda;
+ psrc3 = psrc2 + lda;
+ psrc4 = psrc3 + lda;
+ psrc5 = psrc4 + lda;
+ psrc6 = psrc5 + lda;
+ psrc7 = psrc6 + lda;
+ psrc8 = psrc7 + lda;
+ psrc0 += 8 * lda;
+
+ for (i = (m >> 3); i--;)
+ {
+ LD_DP2(psrc1, 2, src0, src1);
+ LD_DP2(psrc2, 2, src2, src3);
+ LD_DP2(psrc3, 2, src4, src5);
+ LD_DP2(psrc4, 2, src6, src7);
+ LD_DP2(psrc5, 2, src8, src9);
+ LD_DP2(psrc6, 2, src10, src11);
+ LD_DP2(psrc7, 2, src12, src13);
+ LD_DP2(psrc8, 2, src14, src15);
+
+ dst0 = (v2f64) __msa_ilvr_d((v2i64) src2, (v2i64) src0);
+ dst1 = (v2f64) __msa_ilvr_d((v2i64) src6, (v2i64) src4);
+ dst2 = (v2f64) __msa_ilvr_d((v2i64) src10, (v2i64) src8);
+ dst3 = (v2f64) __msa_ilvr_d((v2i64) src14, (v2i64) src12);
+ dst4 = (v2f64) __msa_ilvl_d((v2i64) src2, (v2i64) src0);
+ dst5 = (v2f64) __msa_ilvl_d((v2i64) src6, (v2i64) src4);
+ dst6 = (v2f64) __msa_ilvl_d((v2i64) src10, (v2i64) src8);
+ dst7 = (v2f64) __msa_ilvl_d((v2i64) src14, (v2i64) src12);
+
+ ST_DP8(dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, pdst, 2);
+
+ dst0 = (v2f64) __msa_ilvr_d((v2i64) src3, (v2i64) src1);
+ dst1 = (v2f64) __msa_ilvr_d((v2i64) src7, (v2i64) src5);
+ dst2 = (v2f64) __msa_ilvr_d((v2i64) src11, (v2i64) src9);
+ dst3 = (v2f64) __msa_ilvr_d((v2i64) src15, (v2i64) src13);
+ dst4 = (v2f64) __msa_ilvl_d((v2i64) src3, (v2i64) src1);
+ dst5 = (v2f64) __msa_ilvl_d((v2i64) src7, (v2i64) src5);
+ dst6 = (v2f64) __msa_ilvl_d((v2i64) src11, (v2i64) src9);
+ dst7 = (v2f64) __msa_ilvl_d((v2i64) src15, (v2i64) src13);
+
+ ST_DP8(dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, pdst + 16,
+ 2);
+
+ LD_DP2(psrc1 + 4, 2, src0, src1);
+ LD_DP2(psrc2 + 4, 2, src2, src3);
+ LD_DP2(psrc3 + 4, 2, src4, src5);
+ LD_DP2(psrc4 + 4, 2, src6, src7);
+ LD_DP2(psrc5 + 4, 2, src8, src9);
+ LD_DP2(psrc6 + 4, 2, src10, src11);
+ LD_DP2(psrc7 + 4, 2, src12, src13);
+ LD_DP2(psrc8 + 4, 2, src14, src15);
+
+ dst0 = (v2f64) __msa_ilvr_d((v2i64) src2, (v2i64) src0);
+ dst1 = (v2f64) __msa_ilvr_d((v2i64) src6, (v2i64) src4);
+ dst2 = (v2f64) __msa_ilvr_d((v2i64) src10, (v2i64) src8);
+ dst3 = (v2f64) __msa_ilvr_d((v2i64) src14, (v2i64) src12);
+ dst4 = (v2f64) __msa_ilvl_d((v2i64) src2, (v2i64) src0);
+ dst5 = (v2f64) __msa_ilvl_d((v2i64) src6, (v2i64) src4);
+ dst6 = (v2f64) __msa_ilvl_d((v2i64) src10, (v2i64) src8);
+ dst7 = (v2f64) __msa_ilvl_d((v2i64) src14, (v2i64) src12);
+
+ ST_DP8(dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, pdst + 32,
+ 2);
+
+ dst0 = (v2f64) __msa_ilvr_d((v2i64) src3, (v2i64) src1);
+ dst1 = (v2f64) __msa_ilvr_d((v2i64) src7, (v2i64) src5);
+ dst2 = (v2f64) __msa_ilvr_d((v2i64) src11, (v2i64) src9);
+ dst3 = (v2f64) __msa_ilvr_d((v2i64) src15, (v2i64) src13);
+ dst4 = (v2f64) __msa_ilvl_d((v2i64) src3, (v2i64) src1);
+ dst5 = (v2f64) __msa_ilvl_d((v2i64) src7, (v2i64) src5);
+ dst6 = (v2f64) __msa_ilvl_d((v2i64) src11, (v2i64) src9);
+ dst7 = (v2f64) __msa_ilvl_d((v2i64) src15, (v2i64) src13);
+
+ ST_DP8(dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, pdst + 48,
+ 2);
+
+ psrc1 += 8;
+ psrc2 += 8;
+ psrc3 += 8;
+ psrc4 += 8;
+ psrc5 += 8;
+ psrc6 += 8;
+ psrc7 += 8;
+ psrc8 += 8;
+ pdst += 64;
+ }
+
+ for (i = (m & 7); i--;)
+ {
+ *pdst++ = *psrc1++;
+ *pdst++ = *psrc2++;
+ *pdst++ = *psrc3++;
+ *pdst++ = *psrc4++;
+ *pdst++ = *psrc5++;
+ *pdst++ = *psrc6++;
+ *pdst++ = *psrc7++;
+ *pdst++ = *psrc8++;
+ }
+ }
+
+ if (n & 4)
+ {
+ psrc1 = psrc0;
+ psrc2 = psrc1 + lda;
+ psrc3 = psrc2 + lda;
+ psrc4 = psrc3 + lda;
+ psrc0 += 4 * lda;
+
+ for (i = (m >> 2); i--;)
+ {
+ LD_DP2(psrc1, 2, src0, src1);
+ LD_DP2(psrc2, 2, src2, src3);
+ LD_DP2(psrc3, 2, src4, src5);
+ LD_DP2(psrc4, 2, src6, src7);
+ psrc1 += 4;
+ psrc2 += 4;
+ psrc3 += 4;
+ psrc4 += 4;
+
+ dst0 = (v2f64) __msa_ilvr_d((v2i64) src2, (v2i64) src0);
+ dst1 = (v2f64) __msa_ilvr_d((v2i64) src6, (v2i64) src4);
+ dst2 = (v2f64) __msa_ilvr_d((v2i64) src3, (v2i64) src1);
+ dst3 = (v2f64) __msa_ilvr_d((v2i64) src7, (v2i64) src5);
+
+ dst4 = (v2f64) __msa_ilvl_d((v2i64) src2, (v2i64) src0);
+ dst5 = (v2f64) __msa_ilvl_d((v2i64) src6, (v2i64) src4);
+ dst6 = (v2f64) __msa_ilvl_d((v2i64) src3, (v2i64) src1);
+ dst7 = (v2f64) __msa_ilvl_d((v2i64) src7, (v2i64) src5);
+
+ ST_DP8(dst0, dst1, dst4, dst5, dst2, dst3, dst6, dst7, pdst, 2);
+ pdst += 16;
+ }
+
+ for (i = (m & 3); i--;)
+ {
+ *pdst++ = *psrc1++;
+ *pdst++ = *psrc2++;
+ *pdst++ = *psrc3++;
+ *pdst++ = *psrc4++;
+ }
+ }
+
+ if (n & 2)
+ {
+ psrc1 = psrc0;
+ psrc2 = psrc1 + lda;
+ psrc0 += 2 * lda;
+
+ for (i = (m >> 1); i--;)
+ {
+ src0 = LD_DP(psrc1);
+ src1 = LD_DP(psrc2);
+ psrc1 += 2;
+ psrc2 += 2;
+
+ dst0 = (v2f64) __msa_ilvr_d((v2i64) src1, (v2i64) src0);
+ dst1 = (v2f64) __msa_ilvl_d((v2i64) src1, (v2i64) src0);
+
+ ST_DP2(dst0, dst1, pdst, 2);
+ pdst += 4;
+ }
+
+ if (m & 1)
+ {
+ *pdst++ = *psrc1++;
+ *pdst++ = *psrc2++;
+ }
+ }
+
+ if (n & 1)
+ {
+ psrc1 = psrc0;
+
+ for (i = m; i--;)
+ {
+ *pdst++ = *psrc1++;
+ }
+ }
+
+ return 0;
+}
diff --git a/kernel/mips/dgemm_tcopy_4_msa.c b/kernel/mips/dgemm_tcopy_4_msa.c
new file mode 100644
index 000000000..f147d190e
--- /dev/null
+++ b/kernel/mips/dgemm_tcopy_4_msa.c
@@ -0,0 +1,162 @@
+/*******************************************************************************
+Copyright (c) 2016, 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.
+*******************************************************************************/
+
+#include "common.h"
+#include "macros_msa.h"
+
+int CNAME(BLASLONG m, BLASLONG n, FLOAT * __restrict src, BLASLONG lda,
+ FLOAT * __restrict dst)
+{
+ BLASLONG i, j;
+ FLOAT *psrc0, *psrc1, *psrc2, *psrc3, *psrc4;
+ FLOAT *pdst0, *pdst1, *pdst2, *pdst3;
+ v2f64 src0, src1, src2, src3, src4, src5, src6, src7;
+
+ psrc0 = src;
+ pdst0 = dst;
+
+ pdst2 = dst + m * (n & ~3);
+ pdst3 = dst + m * (n & ~1);
+
+ for (j = (m >> 2); j--;)
+ {
+ psrc1 = psrc0;
+ psrc2 = psrc1 + lda;
+ psrc3 = psrc2 + lda;
+ psrc4 = psrc3 + lda;
+ psrc0 += 4 * lda;
+
+ pdst1 = pdst0;
+ pdst0 += 16;
+
+ for (i = (n >> 2); i--;)
+ {
+ LD_DP2(psrc1, 2, src0, src1);
+ LD_DP2(psrc2, 2, src2, src3);
+ LD_DP2(psrc3, 2, src4, src5);
+ LD_DP2(psrc4, 2, src6, src7);
+ psrc1 += 4;
+ psrc2 += 4;
+ psrc3 += 4;
+ psrc4 += 4;
+
+ ST_DP8(src0, src1, src2, src3, src4, src5, src6, src7, pdst1, 2);
+ pdst1 += m * 4;
+ }
+
+ if (n & 2)
+ {
+ src0 = LD_DP(psrc1);
+ src1 = LD_DP(psrc2);
+ src2 = LD_DP(psrc3);
+ src3 = LD_DP(psrc4);
+ psrc1 += 2;
+ psrc2 += 2;
+ psrc3 += 2;
+ psrc4 += 2;
+
+ ST_DP4(src0, src1, src2, src3, pdst2, 2);
+ pdst2 += 8;
+ }
+
+ if (n & 1)
+ {
+ *pdst3++ = *psrc1++;
+ *pdst3++ = *psrc2++;
+ *pdst3++ = *psrc3++;
+ *pdst3++ = *psrc4++;
+ }
+ }
+
+ if (m & 2)
+ {
+ psrc1 = psrc0;
+ psrc2 = psrc1 + lda;
+ psrc0 += 2 * lda;
+
+ pdst1 = pdst0;
+ pdst0 += 8;
+
+ for (i = (n >> 2); i--;)
+ {
+ LD_DP2(psrc1, 2, src0, src1);
+ LD_DP2(psrc2, 2, src2, src3);
+ psrc1 += 4;
+ psrc2 += 4;
+
+ ST_DP4(src0, src1, src2, src3, pdst1, 2);
+ pdst1 += m * 4;
+ }
+
+ if (n & 2)
+ {
+ src0 = LD_DP(psrc1);
+ src1 = LD_DP(psrc2);
+ psrc1 += 2;
+ psrc2 += 2;
+
+ ST_DP2(src0, src1, pdst2, 2);
+ pdst2 += 4;
+ }
+
+ if (n & 1)
+ {
+ *pdst3++ = *psrc1++;
+ *pdst3++ = *psrc2++;
+ }
+ }
+
+ if (m & 1)
+ {
+ psrc1 = psrc0;
+ pdst1 = pdst0;
+
+ for (i = (n >> 2); i--;)
+ {
+ LD_DP2(psrc1, 2, src0, src1);
+ psrc1 += 4;
+
+ ST_DP2(src0, src1, pdst1, 2);
+ pdst1 += 4 * m;
+ }
+
+ if (n & 2)
+ {
+ src0 = LD_DP(psrc1);
+ psrc1 += 2;
+
+ ST_DP(src0, pdst2);
+ }
+
+ if (n & 1)
+ {
+ *pdst3 = *psrc1;
+ }
+ }
+
+ return 0;
+}
diff --git a/kernel/mips/dgemm_tcopy_8_msa.c b/kernel/mips/dgemm_tcopy_8_msa.c
new file mode 100644
index 000000000..d1ac49b5a
--- /dev/null
+++ b/kernel/mips/dgemm_tcopy_8_msa.c
@@ -0,0 +1,317 @@
+/*******************************************************************************
+Copyright (c) 2016, 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.
+*******************************************************************************/
+
+#include "common.h"
+#include "macros_msa.h"
+
+int CNAME(BLASLONG m, BLASLONG n, FLOAT * __restrict src, BLASLONG lda,
+ FLOAT * __restrict dst)
+{
+ BLASLONG i, j;
+ FLOAT *psrc0, *psrc1, *psrc2, *psrc3, *psrc4;
+ FLOAT *psrc5, *psrc6, *psrc7, *psrc8;
+ FLOAT *pdst0, *pdst1, *pdst2, *pdst3, *pdst4;
+ v2f64 src0, src1, src2, src3, src4, src5, src6, src7;
+ v2f64 src8, src9, src10, src11, src12, src13, src14, src15;
+
+ psrc0 = src;
+ pdst0 = dst;
+
+ pdst2 = dst + m * (n & ~7);
+ pdst3 = dst + m * (n & ~3);
+ pdst4 = dst + m * (n & ~1);
+
+ for (j = (m >> 3); j--;)
+ {
+ psrc1 = psrc0;
+ psrc2 = psrc1 + lda;
+ psrc3 = psrc2 + lda;
+ psrc4 = psrc3 + lda;
+ psrc5 = psrc4 + lda;
+ psrc6 = psrc5 + lda;
+ psrc7 = psrc6 + lda;
+ psrc8 = psrc7 + lda;
+ psrc0 += 8 * lda;
+
+ pdst1 = pdst0;
+ pdst0 += 64;
+
+ for (i = (n >> 3); i--;)
+ {
+ LD_DP4(psrc1, 2, src0, src1, src2, src3);
+ LD_DP4(psrc2, 2, src4, src5, src6, src7);
+ LD_DP4(psrc3, 2, src8, src9, src10, src11);
+ LD_DP4(psrc4, 2, src12, src13, src14, src15);
+ psrc1 += 8;
+ psrc2 += 8;
+ psrc3 += 8;
+ psrc4 += 8;
+
+ ST_DP8(src0, src1, src2, src3, src4, src5, src6, src7, pdst1, 2);
+ ST_DP8(src8, src9, src10, src11, src12, src13, src14, src15,
+ pdst1 + 16, 2);
+
+ LD_DP4(psrc5, 2, src0, src1, src2, src3);
+ LD_DP4(psrc6, 2, src4, src5, src6, src7);
+ LD_DP4(psrc7, 2, src8, src9, src10, src11);
+ LD_DP4(psrc8, 2, src12, src13, src14, src15);
+ psrc5 += 8;
+ psrc6 += 8;
+ psrc7 += 8;
+ psrc8 += 8;
+
+ ST_DP8(src0, src1, src2, src3, src4, src5, src6, src7, pdst1 + 32,
+ 2);
+ ST_DP8(src8, src9, src10, src11, src12, src13, src14, src15,
+ pdst1 + 48, 2);
+ pdst1 += m * 8;
+ }
+
+ if (n & 4)
+ {
+ LD_DP2(psrc1, 2, src0, src1);
+ LD_DP2(psrc2, 2, src2, src3);
+ LD_DP2(psrc3, 2, src4, src5);
+ LD_DP2(psrc4, 2, src6, src7);
+ LD_DP2(psrc5, 2, src8, src9);
+ LD_DP2(psrc6, 2, src10, src11);
+ LD_DP2(psrc7, 2, src12, src13);
+ LD_DP2(psrc8, 2, src14, src15);
+ psrc1 += 4;
+ psrc2 += 4;
+ psrc3 += 4;
+ psrc4 += 4;
+ psrc5 += 4;
+ psrc6 += 4;
+ psrc7 += 4;
+ psrc8 += 4;
+
+ ST_DP8(src0, src1, src2, src3, src4, src5, src6, src7, pdst2, 2);
+ ST_DP8(src8, src9, src10, src11, src12, src13, src14, src15,
+ pdst2 + 16, 2);
+ pdst2 += 32;
+ }
+
+ if (n & 2)
+ {
+ src0 = LD_DP(psrc1);
+ src1 = LD_DP(psrc2);
+ src2 = LD_DP(psrc3);
+ src3 = LD_DP(psrc4);
+ src4 = LD_DP(psrc5);
+ src5 = LD_DP(psrc6);
+ src6 = LD_DP(psrc7);
+ src7 = LD_DP(psrc8);
+ psrc1 += 2;
+ psrc2 += 2;
+ psrc3 += 2;
+ psrc4 += 2;
+ psrc5 += 2;
+ psrc6 += 2;
+ psrc7 += 2;
+ psrc8 += 2;
+
+ ST_DP8(src0, src1, src2, src3, src4, src5, src6, src7, pdst3, 2);
+ pdst3 += 16;
+ }
+
+ if (n & 1)
+ {
+ *pdst4++ = *psrc1++;
+ *pdst4++ = *psrc2++;
+ *pdst4++ = *psrc3++;
+ *pdst4++ = *psrc4++;
+ *pdst4++ = *psrc5++;
+ *pdst4++ = *psrc6++;
+ *pdst4++ = *psrc7++;
+ *pdst4++ = *psrc8++;
+ }
+ }
+
+ if (m & 4)
+ {
+ psrc1 = psrc0;
+ psrc2 = psrc1 + lda;
+ psrc3 = psrc2 + lda;
+ psrc4 = psrc3 + lda;
+ psrc0 += 4 * lda;
+
+ pdst1 = pdst0;
+ pdst0 += 32;
+
+ for (i = (n >> 3); i--;)
+ {
+ LD_DP4(psrc1, 2, src0, src1, src2, src3);
+ LD_DP4(psrc2, 2, src4, src5, src6, src7);
+ LD_DP4(psrc3, 2, src8, src9, src10, src11);
+ LD_DP4(psrc4, 2, src12, src13, src14, src15);
+ psrc1 += 8;
+ psrc2 += 8;
+ psrc3 += 8;
+ psrc4 += 8;
+ psrc5 += 8;
+ psrc6 += 8;
+ psrc7 += 8;
+ psrc8 += 8;
+
+ ST_DP8(src0, src1, src2, src3, src4, src5, src6, src7, pdst1, 2);
+ ST_DP8(src8, src9, src10, src11, src12, src13, src14, src15,
+ pdst1 + 16, 2);
+ pdst1 += 8 * m;
+ }
+
+ if (n & 4)
+ {
+ LD_DP2(psrc1, 2, src0, src1);
+ LD_DP2(psrc2, 2, src2, src3);
+ LD_DP2(psrc3, 2, src4, src5);
+ LD_DP2(psrc4, 2, src6, src7);
+ psrc1 += 4;
+ psrc2 += 4;
+ psrc3 += 4;
+ psrc4 += 4;
+
+ ST_DP8(src0, src1, src2, src3, src4, src5, src6, src7, pdst2, 2);
+ pdst2 += 16;
+ }
+
+ if (n & 2)
+ {
+ src0 = LD_DP(psrc1);
+ src1 = LD_DP(psrc2);
+ src2 = LD_DP(psrc3);
+ src3 = LD_DP(psrc4);
+ psrc1 += 2;
+ psrc2 += 2;
+ psrc3 += 2;
+ psrc4 += 2;
+
+ ST_DP4(src0, src1, src2, src3, pdst3, 2);
+ pdst3 += 8;
+ }
+
+ if (n & 1)
+ {
+ *pdst4++ = *psrc1++;
+ *pdst4++ = *psrc2++;
+ *pdst4++ = *psrc3++;
+ *pdst4++ = *psrc4++;
+ }
+ }
+
+ if (m & 2)
+ {
+ psrc1 = psrc0;
+ psrc2 = psrc1 + lda;
+ psrc0 += 2 * lda;
+
+ pdst1 = pdst0;
+ pdst0 += 16;
+
+ for (i = (n >> 3); i--;)
+ {
+ LD_DP4(psrc1, 2, src0, src1, src2, src3);
+ LD_DP4(psrc2, 2, src4, src5, src6, src7);
+ psrc1 += 8;
+ psrc2 += 8;
+
+ ST_DP8(src0, src1, src2, src3, src4, src5, src6, src7, pdst1, 2);
+ pdst1 += 8 * m;
+ }
+
+ if (n & 4)
+ {
+ LD_DP2(psrc1, 2, src0, src1);
+ LD_DP2(psrc2, 2, src2, src3);
+ psrc1 += 4;
+ psrc2 += 4;
+
+ ST_DP4(src0, src1, src2, src3, pdst2, 2);
+ pdst2 += 8;
+ }
+
+ if (n & 2)
+ {
+ src0 = LD_DP(psrc1);
+ src1 = LD_DP(psrc2);
+ psrc1 += 2;
+ psrc2 += 2;
+
+ ST_DP2(src0, src1, pdst3, 2);
+ pdst3 += 4;
+ }
+
+ if (n & 1)
+ {
+ *pdst4++ = *psrc1++;
+ *pdst4++ = *psrc2++;
+ }
+ }
+
+ if (m & 1)
+ {
+ psrc1 = psrc0;
+ psrc0 += lda;
+
+ pdst1 = pdst0;
+ pdst0 += 8;
+
+ for (i = (n >> 3); i--;)
+ {
+ LD_DP4(psrc1, 2, src0, src1, src2, src3);
+ psrc1 += 8;
+
+ ST_DP4(src0, src1, src2, src3, pdst1, 2);
+ pdst1 += 8 * m;
+ }
+
+ if (n & 4)
+ {
+ LD_DP2(psrc1, 2, src0, src1);
+ psrc1 += 4;
+
+ ST_DP2(src0, src1, pdst2, 2);
+ pdst2 += 4;
+ }
+
+ if (n & 2)
+ {
+ src0 = LD_DP(psrc1);
+ psrc1 += 2;
+
+ ST_DP(src0, pdst3);
+ pdst3 += 2;
+ }
+
+ if (n & 1)
+ {
+ *pdst4++ = *psrc1++;
+ }
+ }
+
+ return 0;
+}
diff --git a/kernel/mips/macros_msa.h b/kernel/mips/macros_msa.h
new file mode 100644
index 000000000..3bcc59629
--- /dev/null
+++ b/kernel/mips/macros_msa.h
@@ -0,0 +1,79 @@
+/*******************************************************************************
+Copyright (c) 2016, 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.
+*******************************************************************************/
+
+#ifndef __MACROS_MSA_H__
+#define __MACROS_MSA_H__
+
+#include
+
+#define LD_D(RTYPE, psrc) *((RTYPE *)(psrc))
+#define LD_DP(...) LD_D(v2f64, __VA_ARGS__)
+
+#define ST_D(RTYPE, in, pdst) *((RTYPE *)(pdst)) = (in)
+#define ST_DP(...) ST_D(v2f64, __VA_ARGS__)
+
+/* Description : Load 2 vectors of double precision floating point elements with stride
+ Arguments : Inputs - psrc, stride
+ Outputs - out0, out1
+ Return Type - double precision floating point
+*/
+#define LD_DP2(psrc, stride, out0, out1) \
+{ \
+ out0 = LD_DP((psrc)); \
+ out1 = LD_DP((psrc) + stride); \
+}
+
+#define LD_DP4(psrc, stride, out0, out1, out2, out3) \
+{ \
+ LD_DP2(psrc, stride, out0, out1) \
+ LD_DP2(psrc + 2 * stride, stride, out2, out3) \
+}
+
+/* Description : Store vectors of double precision floating point elements with stride
+ Arguments : Inputs - in0, in1, pdst, stride
+ Details : Store 2 double precision floating point elements from 'in0' to (pdst)
+ Store 2 double precision floating point elements from 'in1' to (pdst + stride)
+*/
+#define ST_DP2(in0, in1, pdst, stride) \
+{ \
+ ST_DP(in0, (pdst)); \
+ ST_DP(in1, (pdst) + stride); \
+}
+
+#define ST_DP4(in0, in1, in2, in3, pdst, stride) \
+{ \
+ ST_DP2(in0, in1, (pdst), stride); \
+ ST_DP2(in2, in3, (pdst) + 2 * stride, stride); \
+}
+
+#define ST_DP8(in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride) \
+{ \
+ ST_DP4(in0, in1, in2, in3, (pdst), stride); \
+ ST_DP4(in4, in5, in6, in7, (pdst) + 4 * stride, stride); \
+}
+
+#endif /* __MACROS_MSA_H__ */
diff --git a/param.h b/param.h
index 93b1220d6..6948e6a76 100644
--- a/param.h
+++ b/param.h
@@ -2185,8 +2185,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define SGEMM_DEFAULT_UNROLL_M 2
#define SGEMM_DEFAULT_UNROLL_N 2
-#define DGEMM_DEFAULT_UNROLL_M 2
-#define DGEMM_DEFAULT_UNROLL_N 2
+#define DGEMM_DEFAULT_UNROLL_M 8
+#define DGEMM_DEFAULT_UNROLL_N 4
#define CGEMM_DEFAULT_UNROLL_M 2
#define CGEMM_DEFAULT_UNROLL_N 2
@@ -2224,8 +2224,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define SGEMM_DEFAULT_UNROLL_M 2
#define SGEMM_DEFAULT_UNROLL_N 2
-#define DGEMM_DEFAULT_UNROLL_M 2
-#define DGEMM_DEFAULT_UNROLL_N 2
+#define DGEMM_DEFAULT_UNROLL_M 8
+#define DGEMM_DEFAULT_UNROLL_N 4
#define CGEMM_DEFAULT_UNROLL_M 2
#define CGEMM_DEFAULT_UNROLL_N 2