x86_64: opt somatcopy_ct with AVX

Author: XiWeiGu

kernel/x86_64/KERNEL

@@ -493,3 +493,4 @@ CSUMKERNEL = zsum_sse.S
 ZSUMKERNEL = zsum_sse2.S
 
 SOMATCOPY_RT = omatcopy_rt.c
+SOMATCOPY_CT = omatcopy_ct.c

kernel/x86_64/omatcopy_ct.c

@@ -0,0 +1,373 @@
+/***************************************************************************
+Copyright (c) 2024, 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"
+
+#ifdef HAVE_AVX
+
+#define COLS_OF_BLOCK 384
+
+/* +r: %0 = src, %1 = dst, %2 = src_ld, %3 = dst_ld, %4 = dst_tmp */
+/* m: %5 = num_rows, %6 = alpha */
+/* xmm15 = alpha */
+#define TRANS_4x4(a1_no,a2_no,a3_no,a4_no,t1_no,t2_no,t3_no,t4_no)\
+  "vunpcklps %%xmm"#a2_no",%%xmm"#a1_no",%%xmm"#t1_no"; vunpckhps %%xmm"#a2_no",%%xmm"#a1_no",%%xmm"#t2_no";"\
+  "vunpcklps %%xmm"#a4_no",%%xmm"#a3_no",%%xmm"#t3_no"; vunpckhps %%xmm"#a4_no",%%xmm"#a3_no",%%xmm"#t4_no";"\
+  "vunpcklpd %%xmm"#t3_no",%%xmm"#t1_no",%%xmm"#a1_no"; vunpckhpd %%xmm"#t3_no",%%xmm"#t1_no",%%xmm"#a2_no";"\
+  "vunpcklpd %%xmm"#t4_no",%%xmm"#t2_no",%%xmm"#a3_no"; vunpckhpd %%xmm"#t4_no",%%xmm"#t2_no",%%xmm"#a4_no";"
+
+#define TRANS_4x8(a1_no,a2_no,a3_no,a4_no,t1_no,t2_no,t3_no,t4_no)\
+  "vunpcklps %%ymm"#a2_no",%%ymm"#a1_no",%%ymm"#t1_no"; vunpckhps %%ymm"#a2_no",%%ymm"#a1_no",%%ymm"#t2_no";"\
+  "vunpcklps %%ymm"#a4_no",%%ymm"#a3_no",%%ymm"#t3_no"; vunpckhps %%ymm"#a4_no",%%ymm"#a3_no",%%ymm"#t4_no";"\
+  "vunpcklpd %%ymm"#t3_no",%%ymm"#t1_no",%%ymm"#a1_no"; vunpckhpd %%ymm"#t3_no",%%ymm"#t1_no",%%ymm"#a2_no";"\
+  "vunpcklpd %%ymm"#t4_no",%%ymm"#t2_no",%%ymm"#a3_no"; vunpckhpd %%ymm"#t4_no",%%ymm"#t2_no",%%ymm"#a4_no";"
+
+#define SAVE_4x4(b1_no,b2_no,b3_no,b4_no)\
+  "vmovups %%xmm"#b1_no",(%4); vmovups %%xmm"#b2_no",(%4,%3,1); leaq (%4,%3,2),%4;"\
+  "vmovups %%xmm"#b3_no",(%4); vmovups %%xmm"#b4_no",(%4,%3,1); leaq (%4,%3,2),%4;"
+
+#define SAVE_4x8(b1_no,b2_no,b3_no,b4_no) SAVE_4x4(b1_no,b2_no,b3_no,b4_no)\
+  "vextractf128 $1,%%ymm"#b1_no",(%4); vextractf128 $1,%%ymm"#b2_no",(%4,%3,1); leaq (%4,%3,2),%4;"\
+  "vextractf128 $1,%%ymm"#b3_no",(%4); vextractf128 $1,%%ymm"#b4_no",(%4,%3,1); leaq (%4,%3,2),%4;"
+
+#define COPY_4x16 "movq %1,%4; addq $16,%1;"\
+  "vmulps (%0),%%ymm15,%%ymm0; vmulps 32(%0),%%ymm15,%%ymm4; vmulps (%0,%2,1),%%ymm15,%%ymm1; vmulps 32(%0,%2,1),%%ymm15,%%ymm5; leaq (%0,%2,2),%0;"\
+  "vmulps (%0),%%ymm15,%%ymm2; vmulps 32(%0),%%ymm15,%%ymm6; vmulps (%0,%2,1),%%ymm15,%%ymm3; vmulps 32(%0,%2,1),%%ymm15,%%ymm7; leaq (%0,%2,2),%0;"\
+  TRANS_4x8(0,1,2,3,8,9,10,11) SAVE_4x8(0,1,2,3)\
+  TRANS_4x8(4,5,6,7,8,9,10,11) SAVE_4x8(4,5,6,7)
+
+#define COPY_4x8 "movq %1,%4; addq $16,%1;"\
+  "vmulps (%0),%%ymm15,%%ymm0; vmulps (%0,%2,1),%%ymm15,%%ymm1; leaq (%0,%2,2),%0;"\
+  "vmulps (%0),%%ymm15,%%ymm2; vmulps (%0,%2,1),%%ymm15,%%ymm3; leaq (%0,%2,2),%0;"\
+  TRANS_4x8(0,1,2,3,8,9,10,11) SAVE_4x8(0,1,2,3)
+
+#define COPY_4x4 "movq %1,%4; addq $16,%1;"\
+  "vmulps (%0),%%xmm15,%%xmm0; vmulps (%0,%2,1),%%xmm15,%%xmm1; leaq (%0,%2,2),%0;"\
+  "vmulps (%0),%%xmm15,%%xmm2; vmulps (%0,%2,1),%%xmm15,%%xmm3; leaq (%0,%2,2),%0;"\
+  TRANS_4x4(0,1,2,3,8,9,10,11) SAVE_4x4(0,1,2,3)
+
+#define COPY_4x2 \
+  "vmovsd (%0),%%xmm0; vmovhpd (%0,%2,1),%%xmm0,%%xmm0; vmulps %%xmm15,%%xmm0,%%xmm0; leaq (%0,%2,2),%0;"\
+  "vmovsd (%0),%%xmm1; vmovhpd (%0,%2,1),%%xmm1,%%xmm1; vmulps %%xmm15,%%xmm1,%%xmm1; leaq (%0,%2,2),%0;"\
+  "vpermilps $216,%%xmm0,%%xmm0; vpermilps $216,%%xmm1,%%xmm1; vunpcklpd %%xmm1,%%xmm0,%%xmm2; vunpckhpd %%xmm1,%%xmm0,%%xmm3;"\
+  "vmovups %%xmm2,(%1); vmovups %%xmm3,(%1,%3,1); addq $16,%1;"
+
+#define COPY_4x1 \
+  "vmovss (%0),%%xmm0; vinsertps $16,(%0,%2,1),%%xmm0,%%xmm0; leaq (%0,%2,2),%0;"\
+  "vinsertps $32,(%0),%%xmm0,%%xmm0; vinsertps $48,(%0,%2,1),%%xmm0,%%xmm0; leaq (%0,%2,2),%0;"\
+  "vmulps %%xmm15,%%xmm0,%%xmm0; vmovups %%xmm0,(%1); addq $16,%1;"
+
+#define SAVE_2x4(c1_no,c2_no,t1_no,t2_no) \
+  "vunpcklps %%xmm"#c2_no",%%xmm"#c1_no",%%xmm"#t1_no"; vmulps %%xmm15,%%xmm"#t1_no",%%xmm"#t1_no";"\
+  "vmovsd %%xmm"#t1_no",(%4); vmovhpd %%xmm"#t1_no",(%4,%3,1); leaq (%4,%3,2),%4;"\
+  "vunpckhps %%xmm"#c2_no",%%xmm"#c1_no",%%xmm"#t2_no"; vmulps %%xmm15,%%xmm"#t2_no",%%xmm"#t2_no";"\
+  "vmovsd %%xmm"#t2_no",(%4); vmovhpd %%xmm"#t2_no",(%4,%3,1); leaq (%4,%3,2),%4;"
+
+#define COPY_2x16 "movq %1,%4; addq $8,%1;"\
+  "vmovups (%0),%%ymm0; vmovups 32(%0),%%ymm2; vmovups (%0,%2,1),%%ymm1; vmovups 32(%0,%2,1),%%ymm3; leaq (%0,%2,2),%0;"\
+  "vextractf128 $1,%%ymm0,%%xmm4; vextractf128 $1,%%ymm2,%%xmm6; vextractf128 $1,%%ymm1,%%xmm5; vextractf128 $1,%%ymm3,%%xmm7;"\
+  SAVE_2x4(0,1,8,9) SAVE_2x4(4,5,8,9) SAVE_2x4(2,3,8,9) SAVE_2x4(6,7,8,9)
+
+#define COPY_2x8 "movq %1,%4; addq $8,%1;"\
+  "vmovups (%0),%%ymm0; vmovups (%0,%2,1),%%ymm1; leaq (%0,%2,2),%0;"\
+  "vextractf128 $1,%%ymm0,%%xmm2; vextractf128 $1,%%ymm1,%%xmm3;"\
+  SAVE_2x4(0,1,4,5) SAVE_2x4(2,3,4,5)
+
+#define COPY_2x4 "movq %1,%4; addq $8,%1;"\
+  "vmovups (%0),%%xmm0; vmovups (%0,%2,1),%%xmm1; leaq (%0,%2,2),%0;"\
+  SAVE_2x4(0,1,4,5)
+
+#define COPY_2x2 \
+  "vmovsd (%0),%%xmm0; vmovhpd (%0,%2,1),%%xmm0,%%xmm0; vmulps %%xmm15,%%xmm0,%%xmm0; leaq (%0,%2,2),%0; vpermilps $216,%%xmm0,%%xmm0;"\
+  "vmovsd %%xmm0,(%1); vmovhpd %%xmm0,(%1,%3,1); addq $8,%1;"
+
+#define COPY_2x1 \
+  "vmovss (%0),%%xmm0; vinsertps $16,(%0,%2,1),%%xmm0,%%xmm0; vmulps %%xmm15,%%xmm0,%%xmm0; leaq (%0,%2,2),%0; vmovsd %%xmm0,(%1); addq $8,%1;"
+
+#define SAVE_1x4(c1_no)\
+  "vmulps %%xmm15,%%xmm"#c1_no",%%xmm"#c1_no"; vmovss %%xmm"#c1_no",(%4); vextractps $1,%%xmm"#c1_no",(%4,%3,1); leaq (%4,%3,2),%4;"\
+  "vextractps $2,%%xmm"#c1_no",(%4); vextractps $3,%%xmm"#c1_no",(%4,%3,1); leaq (%4,%3,2),%4;"
+
+#define COPY_1x16 "movq %1,%4; addq $4,%1;"\
+  "vmovups (%0),%%xmm1;" SAVE_1x4(1) "vmovups 16(%0),%%xmm2;" SAVE_1x4(2)\
+  "vmovups 32(%0),%%xmm1;" SAVE_1x4(1) "vmovups 48(%0),%%xmm2;" SAVE_1x4(2) "addq %2,%0;"
+
+#define COPY_1x8 "movq %1,%4; addq $4,%1;"\
+  "vmovups (%0),%%xmm1;" SAVE_1x4(1) "vmovups 16(%0),%%xmm2;" SAVE_1x4(2) "addq %2,%0;"
+
+#define COPY_1x4 "movq %1,%4; addq $4,%1; vmovups (%0),%%xmm1;" SAVE_1x4(1) "addq %2,%0;"
+
+#define COPY_1x2 "vmovsd (%0),%%xmm1; addq %2,%0; vmulps %%xmm15,%%xmm1,%%xmm1; vmovss %%xmm1,(%1); vextractps $1,%%xmm1,(%1,%3,1); addq $4,%1;"
+
+#define COPY_1x1 "vmulss (%0),%%xmm15,%%xmm1; vmovss %%xmm1,(%1); addq %2,%0; addq $4,%1;"
+
+#define COMPUTE(ndim){\
+  src = src_base; dst = dst_base;\
+  __asm__ __volatile__(\
+    "vbroadcastss %6,%%ymm15; movq %5,%%r11; cmpq $4,%%r11; jb "#ndim"32f;"\
+    #ndim"31:\n\t"\
+    COPY_4x##ndim "subq $4,%%r11; cmpq $4,%%r11; jnb "#ndim"31b;"\
+    #ndim"32:\n\t"\
+    "cmpq $2,%%r11; jb "#ndim"33f;"\
+    COPY_2x##ndim "subq $2,%%r11;"\
+    #ndim"33:\n\t"\
+    "testq %%r11,%%r11; jz "#ndim"34f;"\
+    COPY_1x##ndim "subq $1,%%r11;"\
+    #ndim"34:\n\t"\
+    :"+r"(src),"+r"(dst),"+r"(src_ld_bytes),"+r"(dst_ld_bytes),"+r"(dst_tmp):"m"(num_cols),"m"(ALPHA):"r11","cc","memory"\
+    ,"xmm0","xmm1","xmm2","xmm3","xmm4","xmm5","xmm6","xmm7","xmm8","xmm9","xmm10","xmm11","xmm12","xmm13","xmm14","xmm15");\
+}
+int CNAME(BLASLONG rows, BLASLONG cols, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG ldb){
+  float *src, *dst, *dst_tmp=0, *src_base, *dst_base;
+  uint64_t src_ld_bytes = (uint64_t)lda * sizeof(float), dst_ld_bytes = (uint64_t)ldb * sizeof(float), num_cols = 0;
+  BLASLONG rows_left, cols_done; float ALPHA = alpha;
+  if(ALPHA==0.0){
+    dst_base = b;
+    for(rows_left=rows;rows_left>0;rows_left--) {memset(dst_base,0,cols*sizeof(float)); dst_base += ldb;}
+    return 0;
+  }
+  for(cols_done=0;cols_done<cols;cols_done+=num_cols){
+    num_cols = cols-cols_done;
+    if(num_cols > COLS_OF_BLOCK) num_cols = COLS_OF_BLOCK;
+    rows_left = rows; src_base = a + (int64_t)lda * (int64_t)cols_done; dst_base = b + cols_done;
+    if(ldb%1024>3 && ldb%1024<1021) for(;rows_left>15;rows_left-=16){COMPUTE(16) src_base += 16; dst_base += 16 * ldb;}
+    for(;rows_left>7;rows_left-=8){COMPUTE(8) src_base += 8; dst_base += 8 * ldb;}
+    for(;rows_left>3;rows_left-=4){COMPUTE(4) src_base += 4; dst_base += 4 * ldb;}
+    for(;rows_left>1;rows_left-=2){COMPUTE(2) src_base += 2; dst_base += 2 * ldb;}
+    if(rows_left>0){COMPUTE(1) src_base ++; dst_base += ldb;}
+  }
+  return 0;
+}
+
+#else
+
+int CNAME(BLASLONG rows, BLASLONG cols, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG ldb)
+{
+    BLASLONG i, j;
+    FLOAT *a_offset, *a_offset1, *a_offset2, *a_offset3, *a_offset4;
+    FLOAT *b_offset, *b_offset1, *b_offset2, *b_offset3, *b_offset4;
+
+    if (rows <= 0)  return 0;
+    if (cols <= 0)  return 0;
+
+    a_offset = a;
+    b_offset = b;
+
+    i = (cols >> 2);
+    if (i > 0) {
+        do {
+            a_offset1 = a_offset;
+            a_offset2 = a_offset1 + lda;
+            a_offset3 = a_offset2 + lda;
+            a_offset4 = a_offset3 + lda;
+            a_offset += 4 * lda;
+
+            b_offset1 = b_offset;
+            b_offset2 = b_offset1 + ldb;
+            b_offset3 = b_offset2 + ldb;
+            b_offset4 = b_offset3 + ldb;
+            b_offset += 4;
+
+            j = (rows >> 2);
+            if (j > 0) {
+                do {
+                /* Column 1 of MAT_B */
+                *(b_offset1 + 0) = *(a_offset1 + 0)*alpha; // Row 1 of MAT_A
+                *(b_offset2 + 0) = *(a_offset1 + 1)*alpha;
+                *(b_offset3 + 0) = *(a_offset1 + 2)*alpha;
+                *(b_offset4 + 0) = *(a_offset1 + 3)*alpha;
+
+                /* Column 2 of MAT_B */
+                *(b_offset1 + 1) = *(a_offset2 + 0)*alpha; // Row 2 of MAT_A
+                *(b_offset2 + 1) = *(a_offset2 + 1)*alpha;
+                *(b_offset3 + 1) = *(a_offset2 + 2)*alpha;
+                *(b_offset4 + 1) = *(a_offset2 + 3)*alpha;
+
+                /* Column 3 of MAT_B */
+                *(b_offset1 + 2) = *(a_offset3 + 0)*alpha; // Row 3 of MAT_A
+                *(b_offset2 + 2) = *(a_offset3 + 1)*alpha;
+                *(b_offset3 + 2) = *(a_offset3 + 2)*alpha;
+                *(b_offset4 + 2) = *(a_offset3 + 3)*alpha;
+
+                /* Column 4 of MAT_B */
+                *(b_offset1 + 3) = *(a_offset4 + 0)*alpha; // Row 4 of MAT_A
+                *(b_offset2 + 3) = *(a_offset4 + 1)*alpha;
+                *(b_offset3 + 3) = *(a_offset4 + 2)*alpha;
+                *(b_offset4 + 3) = *(a_offset4 + 3)*alpha;
+
+                a_offset1 += 4;
+                a_offset2 += 4;
+                a_offset3 += 4;
+                a_offset4 += 4;
+                b_offset1 += ldb * 4;
+                b_offset2 += ldb * 4;
+                b_offset3 += ldb * 4;
+                b_offset4 += ldb * 4;
+
+                j--;
+                } while (j > 0);
+            } // if(j > 0)
+
+
+            if (rows & 2) {
+                *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+                *(b_offset2 + 0) = *(a_offset1 + 1)*alpha;
+
+                *(b_offset1 + 1) = *(a_offset2 + 0)*alpha;
+                *(b_offset2 + 1) = *(a_offset2 + 1)*alpha;
+
+                *(b_offset1 + 2) = *(a_offset3 + 0)*alpha;
+                *(b_offset2 + 2) = *(a_offset3 + 1)*alpha;
+
+                *(b_offset1 + 3) = *(a_offset4 + 0)*alpha;
+                *(b_offset2 + 3) = *(a_offset4 + 1)*alpha;
+
+                a_offset1 += 2;
+                a_offset2 += 2;
+                a_offset3 += 2;
+                a_offset4 += 2;
+
+                b_offset1 += ldb*2;
+
+            }
+
+            if (rows & 1) {
+                *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+
+                *(b_offset1 + 1) = *(a_offset2 + 0)*alpha;
+
+                *(b_offset1 + 2) = *(a_offset3 + 0)*alpha;
+
+                *(b_offset1 + 3) = *(a_offset4 + 0)*alpha;
+            }
+
+            i--;
+        } while (i > 0);
+    }
+
+
+    if (cols & 2) {
+        a_offset1 = a_offset;
+        a_offset2 = a_offset1 + lda;
+        a_offset += 2 * lda;
+
+        b_offset1 = b_offset;
+        b_offset2 = b_offset1 + ldb;
+        b_offset3 = b_offset2 + ldb;
+        b_offset4 = b_offset3 + ldb;
+        b_offset += 2;
+
+        j = (rows >> 2);
+        if (j > 0){
+            do {
+                *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+                *(b_offset2 + 0) = *(a_offset1 + 1)*alpha;
+                *(b_offset3 + 0) = *(a_offset1 + 2)*alpha;
+                *(b_offset4 + 0) = *(a_offset1 + 3)*alpha;
+
+                *(b_offset1 + 1) = *(a_offset2 + 0)*alpha;
+                *(b_offset2 + 1) = *(a_offset2 + 1)*alpha;
+                *(b_offset3 + 1) = *(a_offset2 + 2)*alpha;
+                *(b_offset4 + 1) = *(a_offset2 + 3)*alpha;
+
+                a_offset1 += 4;
+                a_offset2 += 4;
+                b_offset1 += ldb * 4;
+                b_offset2 += ldb * 4;
+                b_offset3 += ldb * 4;
+                b_offset4 += ldb * 4;
+
+                j--;
+            } while (j > 0);
+        }
+
+
+        if (rows & 2){
+            *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+            *(b_offset2 + 0) = *(a_offset1 + 1)*alpha;
+
+            *(b_offset1 + 1) = *(a_offset2 + 0)*alpha;
+            *(b_offset2 + 1) = *(a_offset2 + 1)*alpha;
+
+            a_offset1 += 2;
+            a_offset2 += 2;
+            b_offset1 += ldb*2;
+
+        }
+
+
+        if (rows & 1){
+            *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+            *(b_offset1 + 1) = *(a_offset2 + 0)*alpha;
+        }
+    } // if (cols & 2)
+
+
+    if (cols & 1) {
+        a_offset1 = a_offset;
+        a_offset += lda;
+
+        b_offset1 = b_offset;
+        b_offset2 = b_offset1 + ldb;
+        b_offset3 = b_offset2 + ldb;
+        b_offset4 = b_offset3 + ldb;
+
+        j = (rows >> 2);
+        if (j > 0){
+            do {
+                *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+                *(b_offset2 + 0) = *(a_offset1 + 1)*alpha;
+                *(b_offset3 + 0) = *(a_offset1 + 2)*alpha;
+                *(b_offset4 + 0) = *(a_offset1 + 3)*alpha;
+
+                a_offset1 += 4;
+                b_offset1 += ldb * 4;
+                b_offset2 += ldb * 4;
+                b_offset3 += ldb * 4;
+                b_offset4 += ldb * 4;
+
+                j--;
+            } while (j > 0);
+        }
+
+        if (rows & 2){
+            *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+            *(b_offset2 + 0) = *(a_offset1 + 1)*alpha;
+
+            a_offset1 += 2;
+            b_offset1 += ldb * 2;
+        }
+
+        if (rows & 1){
+            *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+        }
+    }
+
+    return 0;
+}
+
+#endif