/* BLIS An object-based framework for developing high-performance BLAS-like libraries. Copyright (C) 2014, The University of Texas at Austin Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: - Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. - 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. - Neither the name(s) of the copyright holder(s) 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 COPYRIGHT HOLDER 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 "blis.h" void bli_sgemmtrsm_u_opt_d4x2( dim_t k, float* restrict alpha, float* restrict a12, float* restrict a11, float* restrict bd21, float* restrict bd11, float* restrict b11, float* restrict c11, inc_t rs_c, inc_t cs_c, float* restrict a_next, float* restrict b_next ) { bli_check_error_code( BLIS_NOT_YET_IMPLEMENTED ); } void bli_dgemmtrsm_u_opt_d4x2( dim_t k, double* restrict alpha, double* restrict a12, double* restrict a11, double* restrict bd21, double* restrict bd11, double* restrict b11, double* restrict c11, inc_t rs_c, inc_t cs_c, double* restrict a_next, double* restrict b_next ) { dim_t k_iter; dim_t k_left; k_iter = k / 8; k_left = k % 8; __asm__ volatile ( " \n\t" "movl %2, %%eax \n\t" // load address of a12. "movl %4, %%ebx \n\t" // load address of bd21. " \n\t" "addl $8 * 16, %%eax \n\t" // increment pointers to allow byte "addl $8 * 16, %%ebx \n\t" // offsets in the unrolled iterations. " \n\t" "movapd -8 * 16(%%eax), %%xmm0 \n\t" // initialize loop by pre-loading elements "movapd -4 * 16(%%eax), %%xmm3 \n\t" // of a. " \n\t" "pxor %%xmm4, %%xmm4 \n\t" "pxor %%xmm5, %%xmm5 \n\t" "pxor %%xmm6, %%xmm6 \n\t" "pxor %%xmm7, %%xmm7 \n\t" " \n\t" " \n\t" " \n\t" "movl %0, %%esi \n\t" // i = k_iter; "testl %%esi, %%esi \n\t" // check i via logical AND. "je .CONSIDERKLEFT \n\t" // if i == 0, jump to code that " \n\t" // contains the k_left loop. " \n\t" " \n\t" ".LOOPKITER: \n\t" // MAIN LOOP " \n\t" "movapd -8 * 16(%%ebx), %%xmm1 \n\t" // iteration 0 "movapd %%xmm1, %%xmm2 \n\t" "mulpd %%xmm0, %%xmm1 \n\t" "addpd %%xmm1, %%xmm4 \n\t" "movapd -7 * 16(%%ebx), %%xmm1 \n\t" "mulpd %%xmm1, %%xmm0 \n\t" "addpd %%xmm0, %%xmm5 \n\t" "movapd -7 * 16(%%eax), %%xmm0 \n\t" "mulpd %%xmm0, %%xmm2 \n\t" "mulpd %%xmm0, %%xmm1 \n\t" "movapd -6 * 16(%%eax), %%xmm0 \n\t" "addpd %%xmm2, %%xmm6 \n\t" "addpd %%xmm1, %%xmm7 \n\t" " \n\t" "movapd -6 * 16(%%ebx), %%xmm1 \n\t" // iteration 1 "movapd %%xmm1, %%xmm2 \n\t" "mulpd %%xmm0, %%xmm1 \n\t" "addpd %%xmm1, %%xmm4 \n\t" "movapd -5 * 16(%%ebx), %%xmm1 \n\t" "mulpd %%xmm1, %%xmm0 \n\t" "addpd %%xmm0, %%xmm5 \n\t" "movapd -5 * 16(%%eax), %%xmm0 \n\t" "mulpd %%xmm0, %%xmm2 \n\t" "mulpd %%xmm0, %%xmm1 \n\t" "movapd 0 * 16(%%eax), %%xmm0 \n\t" "addpd %%xmm2, %%xmm6 \n\t" "addpd %%xmm1, %%xmm7 \n\t" " \n\t" "movapd -4 * 16(%%ebx), %%xmm1 \n\t" // iteration 2 "movapd %%xmm1, %%xmm2 \n\t" "mulpd %%xmm3, %%xmm1 \n\t" "addpd %%xmm1, %%xmm4 \n\t" "movapd -3 * 16(%%ebx), %%xmm1 \n\t" "mulpd %%xmm1, %%xmm3 \n\t" "addpd %%xmm3, %%xmm5 \n\t" "movapd -3 * 16(%%eax), %%xmm3 \n\t" "mulpd %%xmm3, %%xmm2 \n\t" "mulpd %%xmm3, %%xmm1 \n\t" "movapd -2 * 16(%%eax), %%xmm3 \n\t" "addpd %%xmm2, %%xmm6 \n\t" "addpd %%xmm1, %%xmm7 \n\t" " \n\t" "movapd -2 * 16(%%ebx), %%xmm1 \n\t" // iteration 3 "movapd %%xmm1, %%xmm2 \n\t" "mulpd %%xmm3, %%xmm1 \n\t" "addpd %%xmm1, %%xmm4 \n\t" "movapd -1 * 16(%%ebx), %%xmm1 \n\t" "mulpd %%xmm1, %%xmm3 \n\t" "addpd %%xmm3, %%xmm5 \n\t" "movapd -1 * 16(%%eax), %%xmm3 \n\t" "mulpd %%xmm3, %%xmm2 \n\t" "mulpd %%xmm3, %%xmm1 \n\t" "movapd 4 * 16(%%eax), %%xmm3 \n\t" "addpd %%xmm2, %%xmm6 \n\t" "addpd %%xmm1, %%xmm7 \n\t" " \n\t" "movapd 0 * 16(%%ebx), %%xmm1 \n\t" // iteration 4 "movapd %%xmm1, %%xmm2 \n\t" "mulpd %%xmm0, %%xmm1 \n\t" "addpd %%xmm1, %%xmm4 \n\t" "movapd 1 * 16(%%ebx), %%xmm1 \n\t" "mulpd %%xmm1, %%xmm0 \n\t" "addpd %%xmm0, %%xmm5 \n\t" "movapd 1 * 16(%%eax), %%xmm0 \n\t" "mulpd %%xmm0, %%xmm2 \n\t" "mulpd %%xmm0, %%xmm1 \n\t" "movapd 2 * 16(%%eax), %%xmm0 \n\t" "addpd %%xmm2, %%xmm6 \n\t" "addpd %%xmm1, %%xmm7 \n\t" " \n\t" "movapd 2 * 16(%%ebx), %%xmm1 \n\t" // iteration 5 "movapd %%xmm1, %%xmm2 \n\t" "mulpd %%xmm0, %%xmm1 \n\t" "addpd %%xmm1, %%xmm4 \n\t" "movapd 3 * 16(%%ebx), %%xmm1 \n\t" "mulpd %%xmm1, %%xmm0 \n\t" "addpd %%xmm0, %%xmm5 \n\t" "movapd 3 * 16(%%eax), %%xmm0 \n\t" "mulpd %%xmm0, %%xmm2 \n\t" "mulpd %%xmm0, %%xmm1 \n\t" "movapd 8 * 16(%%eax), %%xmm0 \n\t" "addpd %%xmm2, %%xmm6 \n\t" "addpd %%xmm1, %%xmm7 \n\t" " \n\t" "movapd 4 * 16(%%ebx), %%xmm1 \n\t" // iteration 6 "movapd %%xmm1, %%xmm2 \n\t" "mulpd %%xmm3, %%xmm1 \n\t" "addpd %%xmm1, %%xmm4 \n\t" "movapd 5 * 16(%%ebx), %%xmm1 \n\t" "mulpd %%xmm1, %%xmm3 \n\t" "addpd %%xmm3, %%xmm5 \n\t" "movapd 5 * 16(%%eax), %%xmm3 \n\t" "mulpd %%xmm3, %%xmm2 \n\t" "mulpd %%xmm3, %%xmm1 \n\t" "movapd 6 * 16(%%eax), %%xmm3 \n\t" "addl $8 * 4 * 8, %%eax \n\t" // a += 8*4 (unroll x mr) "addpd %%xmm2, %%xmm6 \n\t" "addpd %%xmm1, %%xmm7 \n\t" " \n\t" "movapd 6 * 16(%%ebx), %%xmm1 \n\t" // iteration 7 "movapd %%xmm1, %%xmm2 \n\t" "mulpd %%xmm3, %%xmm1 \n\t" "addpd %%xmm1, %%xmm4 \n\t" "movapd 7 * 16(%%ebx), %%xmm1 \n\t" "addl $8 * 2 * 2 * 8, %%ebx \n\t" // b += 8*2*2 (unroll x nr x ndup) "mulpd %%xmm1, %%xmm3 \n\t" "addpd %%xmm3, %%xmm5 \n\t" "movapd -9 * 16(%%eax), %%xmm3 \n\t" "mulpd %%xmm3, %%xmm2 \n\t" "mulpd %%xmm3, %%xmm1 \n\t" "decl %%esi \n\t" // i -= 1; "movapd -4 * 16(%%eax), %%xmm3 \n\t" "addpd %%xmm2, %%xmm6 \n\t" "addpd %%xmm1, %%xmm7 \n\t" " \n\t" "jne .LOOPKITER \n\t" // iterate again if i != 0. " \n\t" " \n\t" " \n\t" ".CONSIDERKLEFT: \n\t" " \n\t" "movl %1, %%esi \n\t" // i = k_left; "testl %%esi, %%esi \n\t" // check i via logical AND. "je .POSTACCUM \n\t" // if i == 0, we're done; jump to end. " \n\t" // else, we prepare to enter k_left loop. " \n\t" " \n\t" ".LOOPKLEFT: \n\t" // EDGE LOOP " \n\t" "movapd -8 * 16(%%ebx), %%xmm1 \n\t" // iteration i "movapd %%xmm1, %%xmm2 \n\t" "mulpd %%xmm0, %%xmm1 \n\t" "addpd %%xmm1, %%xmm4 \n\t" "movapd -7 * 16(%%ebx), %%xmm1 \n\t" "addl $1 * 2 * 2 * 8, %%ebx \n\t" // b += 2*2 (1 x nr x ndup) "mulpd %%xmm1, %%xmm0 \n\t" "addpd %%xmm0, %%xmm5 \n\t" "movapd -7 * 16(%%eax), %%xmm0 \n\t" "mulpd %%xmm0, %%xmm2 \n\t" "mulpd %%xmm0, %%xmm1 \n\t" "movapd -6 * 16(%%eax), %%xmm0 \n\t" "addl $1 * 4 * 8, %%eax \n\t" // a += 4 (1 x mr) "addpd %%xmm2, %%xmm6 \n\t" "addpd %%xmm1, %%xmm7 \n\t" " \n\t" "decl %%esi \n\t" // i -= 1; "jne .LOOPKLEFT \n\t" // iterate again if i != 0. " \n\t" " \n\t" " \n\t" ".POSTACCUM: \n\t" " \n\t" " \n\t" "movl %6, %%ebx \n\t" // load address of b11. " \n\t" " \n\t" // xmm4 == ( ab00 xmm5 == ( ab01 " \n\t" // ab10 ) ab11 ) " \n\t" // xmm6 == ( ab20 xmm7 == ( ab21 " \n\t" // ab30 ) ab31 ) "movapd %%xmm4, %%xmm0 \n\t" "unpcklpd %%xmm5, %%xmm0 \n\t" "unpckhpd %%xmm5, %%xmm4 \n\t" "movapd %%xmm4, %%xmm1 \n\t" " \n\t" "movapd %%xmm6, %%xmm2 \n\t" "unpcklpd %%xmm7, %%xmm2 \n\t" "unpckhpd %%xmm7, %%xmm6 \n\t" "movapd %%xmm6, %%xmm3 \n\t" " \n\t" // xmm0 == ( ab00 ab01 ) " \n\t" // xmm1 == ( ab10 ab11 ) " \n\t" // xmm2 == ( ab20 ab21 ) " \n\t" // xmm3 == ( ab30 ab31 ) " \n\t" "movl %10, %%eax \n\t" // load address of alpha "movddup (%%eax), %%xmm7 \n\t" // load alpha and duplicate " \n\t" "movapd 0 * 16(%%ebx), %%xmm4 \n\t" // load xmm4 = ( beta00 beta01 ) "movapd 1 * 16(%%ebx), %%xmm5 \n\t" // load xmm5 = ( beta10 beta11 ) "movapd 2 * 16(%%ebx), %%xmm6 \n\t" // load xmm6 = ( beta20 beta21 ) "mulpd %%xmm7, %%xmm4 \n\t" // xmm4 *= alpha "mulpd %%xmm7, %%xmm5 \n\t" // xmm5 *= alpha "mulpd %%xmm7, %%xmm6 \n\t" // xmm6 *= alpha //"movapd 3 * 16(%%ebx), %%xmm7 \n\t" // load xmm7 = ( beta30 beta31 ) "mulpd 3 * 16(%%ebx), %%xmm7 \n\t" // xmm7 = alpha * ( beta30 beta31 ) " \n\t" "subpd %%xmm0, %%xmm4 \n\t" // xmm4 -= xmm0 "subpd %%xmm1, %%xmm5 \n\t" // xmm5 -= xmm1 "subpd %%xmm2, %%xmm6 \n\t" // xmm6 -= xmm2 "subpd %%xmm3, %%xmm7 \n\t" // xmm7 -= xmm3 " \n\t" " \n\t" " \n\t" ".TRSM: \n\t" " \n\t" " \n\t" "movl %3, %%eax \n\t" // load address of a11 "movl %7, %%ecx \n\t" // load address of c11 " \n\t" "movl %8, %%edi \n\t" // load rs_c "movl %9, %%esi \n\t" // load cs_c "sall $3, %%edi \n\t" // rs_c *= sizeof( double ) "sall $3, %%esi \n\t" // cs_c *= sizeof( double ) " \n\t" "addl %%edi, %%ecx \n\t" // c11 += (4-1)*rs_c "addl %%edi, %%ecx \n\t" "addl %%edi, %%ecx \n\t" " \n\t" " \n\t" " \n\t" // iteration 0 " \n\t" "movddup (3+3*4)*8(%%eax), %%xmm3 \n\t" // load xmm3 = (1/alpha33) " \n\t" "mulpd %%xmm3, %%xmm7 \n\t" // xmm7 *= (1/alpha33); " \n\t" "movapd %%xmm7, 3 * 16(%%ebx) \n\t" // store ( beta30 beta31 ) = xmm7 "movlpd %%xmm7, (%%ecx) \n\t" // store ( gamma30 ) = xmm7[0] "movhpd %%xmm7, (%%ecx,%%esi) \n\t" // store ( gamma31 ) = xmm7[1] "subl %%edi, %%ecx \n\t" // c11 -= rs_c " \n\t" " \n\t" " \n\t" " \n\t" // iteration 1 " \n\t" "movddup (2+2*4)*8(%%eax), %%xmm2 \n\t" // load xmm2 = (1/alpha22) "movddup (2+3*4)*8(%%eax), %%xmm3 \n\t" // load xmm3 = alpha23 " \n\t" "mulpd %%xmm7, %%xmm3 \n\t" // xmm3 = alpha23 * ( beta30 beta31 ) "subpd %%xmm3, %%xmm6 \n\t" // xmm6 -= xmm3 "mulpd %%xmm2, %%xmm6 \n\t" // xmm6 *= (1/alpha22); " \n\t" "movapd %%xmm6, 2 * 16(%%ebx) \n\t" // store ( beta20 beta21 ) = xmm6 "movlpd %%xmm6, (%%ecx) \n\t" // store ( gamma20 ) = xmm6[0] "movhpd %%xmm6, (%%ecx,%%esi) \n\t" // store ( gamma21 ) = xmm6[1] "subl %%edi, %%ecx \n\t" // c11 -= rs_c " \n\t" " \n\t" " \n\t" " \n\t" // iteration 2 " \n\t" "movddup (1+1*4)*8(%%eax), %%xmm1 \n\t" // load xmm1 = (1/alpha11) "movddup (1+2*4)*8(%%eax), %%xmm2 \n\t" // load xmm2 = alpha12 "movddup (1+3*4)*8(%%eax), %%xmm3 \n\t" // load xmm3 = alpha13 " \n\t" "mulpd %%xmm6, %%xmm2 \n\t" // xmm2 = alpha12 * ( beta20 beta21 ) "mulpd %%xmm7, %%xmm3 \n\t" // xmm3 = alpha13 * ( beta30 beta31 ) "addpd %%xmm3, %%xmm2 \n\t" // xmm2 += xmm3; "subpd %%xmm2, %%xmm5 \n\t" // xmm5 -= xmm2 "mulpd %%xmm1, %%xmm5 \n\t" // xmm5 *= (1/alpha11); " \n\t" "movapd %%xmm5, 1 * 16(%%ebx) \n\t" // store ( beta10 beta11 ) = xmm5 "movlpd %%xmm5, (%%ecx) \n\t" // store ( gamma10 ) = xmm5[0] "movhpd %%xmm5, (%%ecx,%%esi) \n\t" // store ( gamma11 ) = xmm5[1] "subl %%edi, %%ecx \n\t" // c11 -= rs_c " \n\t" " \n\t" " \n\t" " \n\t" // iteration 3 " \n\t" "movddup (0+0*4)*8(%%eax), %%xmm0 \n\t" // load xmm0 = (1/alpha00) "movddup (0+1*4)*8(%%eax), %%xmm1 \n\t" // load xmm1 = alpha01 "movddup (0+2*4)*8(%%eax), %%xmm2 \n\t" // load xmm2 = alpha02 "movddup (0+3*4)*8(%%eax), %%xmm3 \n\t" // load xmm3 = alpha03 " \n\t" "mulpd %%xmm5, %%xmm1 \n\t" // xmm1 = alpha01 * ( beta10 beta11 ) "mulpd %%xmm6, %%xmm2 \n\t" // xmm2 = alpha02 * ( beta20 beta21 ) "mulpd %%xmm7, %%xmm3 \n\t" // xmm3 = alpha03 * ( beta30 beta31 ) "addpd %%xmm2, %%xmm1 \n\t" // xmm1 += xmm2; "addpd %%xmm3, %%xmm1 \n\t" // xmm1 += xmm3; "subpd %%xmm1, %%xmm4 \n\t" // xmm4 -= xmm1 "mulpd %%xmm0, %%xmm4 \n\t" // xmm4 *= (1/alpha00); " \n\t" "movapd %%xmm4, 0 * 16(%%ebx) \n\t" // store ( beta00 beta01 ) = xmm4 "movlpd %%xmm4, (%%ecx) \n\t" // store ( gamma00 ) = xmm4[0] "movhpd %%xmm4, (%%ecx,%%esi) \n\t" // store ( gamma01 ) = xmm4[1] " \n\t" " \n\t" " \n\t" ".UPDATEBD11: \n\t" " \n\t" " \n\t" "movl %5, %%edx \n\t" " \n\t" "movddup %%xmm4, %%xmm0 \n\t" "movddup %%xmm5, %%xmm1 \n\t" "movddup %%xmm6, %%xmm2 \n\t" "movddup %%xmm7, %%xmm3 \n\t" " \n\t" "unpckhpd %%xmm4, %%xmm4 \n\t" "unpckhpd %%xmm5, %%xmm5 \n\t" "unpckhpd %%xmm6, %%xmm6 \n\t" "unpckhpd %%xmm7, %%xmm7 \n\t" " \n\t" "movapd %%xmm0, 0 * 16(%%edx) \n\t" "movapd %%xmm4, 1 * 16(%%edx) \n\t" "movapd %%xmm1, 2 * 16(%%edx) \n\t" "movapd %%xmm5, 3 * 16(%%edx) \n\t" "movapd %%xmm2, 4 * 16(%%edx) \n\t" "movapd %%xmm6, 5 * 16(%%edx) \n\t" "movapd %%xmm3, 6 * 16(%%edx) \n\t" "movapd %%xmm7, 7 * 16(%%edx) \n\t" " \n\t" " \n\t" : // output operands (none) : // input operands "m" (k_iter), "m" (k_left), "m" (a12), "m" (a11), "m" (bd21), "m" (bd11), "m" (b11), "m" (c11), "m" (rs_c), "m" (cs_c), "m" (alpha) : // register clobber list "eax", "ebx", "ecx", "edx", "esi", "edi", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7", "memory" ); } void bli_cgemmtrsm_u_opt_d4x2( dim_t k, scomplex* restrict alpha, scomplex* restrict a12, scomplex* restrict a11, scomplex* restrict bd21, scomplex* restrict bd11, scomplex* restrict b11, scomplex* restrict c11, inc_t rs_c, inc_t cs_c, scomplex* restrict a_next, scomplex* restrict b_next ) { bli_check_error_code( BLIS_NOT_YET_IMPLEMENTED ); } void bli_zgemmtrsm_u_opt_d4x2( dim_t k, dcomplex* restrict alpha, dcomplex* restrict a12, dcomplex* restrict a11, dcomplex* restrict bd21, dcomplex* restrict bd11, dcomplex* restrict b11, dcomplex* restrict c11, inc_t rs_c, inc_t cs_c, dcomplex* restrict a_next, dcomplex* restrict b_next ) { bli_check_error_code( BLIS_NOT_YET_IMPLEMENTED ); }