/*************************************************************************** Copyright (c) 2013-2018, 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 #if defined(DOUBLE) #define ABS fabs #else #define ABS fabsf #endif /** * Find minimum index * Warning: requirements n>0 and n % 32 == 0 * @param n * @param x pointer to the vector * @param minf (out) minimum absolute value .( only for output ) * @return minimum index */ static BLASLONG diamin_kernel_32(BLASLONG n, FLOAT *x, FLOAT *minf) { BLASLONG index; register __vector long long start = {1,0}; register __vector long long temp_add_index = {2, 2}; __asm__( "lxvd2x 44, 0,%[ptr_tmp] \n\t" "lxvd2x 45, %[i16],%[ptr_tmp] \n\t" "lxvd2x 46, %[i32],%[ptr_tmp] \n\t" "lxvd2x 47, %[i48],%[ptr_tmp] \n\t" "lxvd2x 48, %[i64],%[ptr_tmp] \n\t" "lxvd2x 49, %[i80],%[ptr_tmp] \n\t" "lxvd2x 50, %[i96],%[ptr_tmp] \n\t" "lxvd2x 51,%[i112],%[ptr_tmp] \n\t" "xxlor 40,%x[start],%x[start] \n\t" //{ 1,0} vs40 | v8 "vaddudm 9,8, %[adder] \n\t" //{3,2} vs41 "xxlxor 37,37 ,37 \n\t" //v5 v37 index_count "vaddudm 10,9,%[adder] \n\t" //{5,4} vs42 "xxlxor 38 ,38 ,38 \n\t" // v6 | vs38 vec_min_index "vaddudm 11,10,%[adder] \n\t" //{7,6} vs43 "lxvdsx 39,0,%[ptr_minf] \n\t" // vs39 vec_min_value "vaddudm 4,11, %[adder] \n\t" // {9,8} -{8;8} vs36 | v4 "xxspltd 36,36,0 \n\t" "xvabsdp 39, 39 \n\t" "xvabsdp 44, 44 \n\t" "xvabsdp 45, 45 \n\t" "xvabsdp 46, 46 \n\t" "xvabsdp 47, 47 \n\t" "xvabsdp 48, 48 \n\t" "xvabsdp 49, 49 \n\t" "xvabsdp 50, 50 \n\t" "xvabsdp 51, 51 \n\t" //jump first half forward "b 2f \n\t" //=================================================================== ".p2align 5 \n\t" "1: \n\t" "xvcmpgtdp 2,44,45 \n\t " "xvcmpgtdp 3,46,47 \n\t " "xvcmpgtdp 4,48,49 \n\t " "xvcmpgtdp 5,50,51 \n\t" "xxsel 32,40,41,2 \n\t" "xxsel 0,44,45,2 \n\t" "xxsel 33,42,43,3 \n\t" "xxsel 1,46,47,3 \n\t" "xxsel 34,40,41,4 \n\t" "xxsel 45,48,49,4 \n\t" "xxsel 35,42,43,5 \n\t" "xxsel 47,50,51,5 \n\t" "xvcmpgtdp 2,0, 1 \n\t" "xvcmpgtdp 3, 45,47 \n\t" "addi %[ptr_tmp] ,%[ptr_tmp] , 128 \n\t" "xxsel 32,32,33,2 \n\t" "xxsel 0 ,0,1,2 \n\t" "xxsel 34,34,35,3 \n\t" "xxsel 5,45,47,3 \n\t" //load next 64 "lxvd2x 44, 0,%[ptr_tmp] \n\t" "lxvd2x 45, %[i16],%[ptr_tmp] \n\t" // for {second 8 elements } we have to add 8 to each so that it became {from 8 to 16} "vaddudm 2,2,4 \n\t" // vs34=vs34 + vs36{8,8} "lxvd2x 46, %[i32],%[ptr_tmp] \n\t" "lxvd2x 47, %[i48],%[ptr_tmp] \n\t" //choose smaller from first and second part "xvcmpgtdp 4, 0,5 \n\t" "xxsel 3, 0,5,4 \n\t" "xxsel 33,32,34,4 \n\t" //load next 64 "lxvd2x 48, %[i64],%[ptr_tmp] \n\t" "lxvd2x 49, %[i80],%[ptr_tmp] \n\t" "vaddudm 1,1,5 \n\t" // get real index for first smaller "lxvd2x 50, %[i96],%[ptr_tmp] \n\t" "lxvd2x 51,%[i112],%[ptr_tmp] \n\t" //compare with previous to get vec_min_index(v6 | vs38 ) and vec_min_value (vs39) "xvcmpgtdp 2,39, 3 \n\t" "xxsel 39,39,3,2 \n\t" "xxsel 38,38,33,2 \n\t" //update index += 8 "vaddudm 5,5,4 \n\t" "xvabsdp 44, 44 \n\t" "xvabsdp 45, 45 \n\t" "xvabsdp 46, 46 \n\t" "xvabsdp 47, 47 \n\t" //update index += 8 "vaddudm 5,5,4 \n\t" "xvabsdp 48, 48 \n\t" "xvabsdp 49, 49 \n\t" "xvabsdp 50, 50 \n\t" "xvabsdp 51, 51 \n\t" //<-----------jump here from first load "2: \n\t" "xvcmpgtdp 2,44,45 \n\t " "xvcmpgtdp 3,46,47 \n\t " "xvcmpgtdp 4,48,49 \n\t " "xvcmpgtdp 5,50,51 \n\t" "xxsel 32,40,41,2 \n\t" "xxsel 0,44,45,2 \n\t" "xxsel 33,42,43,3 \n\t" "xxsel 1,46,47,3 \n\t" "xxsel 34,40,41,4 \n\t" "xxsel 45,48,49,4 \n\t" "xxsel 35,42,43,5 \n\t" "xxsel 47,50,51,5 \n\t" "xvcmpgtdp 2,0, 1 \n\t" "xvcmpgtdp 3, 45,47 \n\t" "xxsel 32,32,33,2 \n\t" "xxsel 0 ,0,1,2 \n\t" "xxsel 34,34,35,3 \n\t" "xxsel 5,45,47,3 \n\t" "addi %[ptr_tmp] ,%[ptr_tmp] , 128 \n\t" // for {second 8 elements } we have to add 8 to each so that it became {from 8 to 16} "vaddudm 2,2,4 \n\t" // vs34=vs34 + vs36{8,8} //load next 64 "lxvd2x 44, 0,%[ptr_tmp] \n\t" "lxvd2x 45, %[i16],%[ptr_tmp] \n\t" "lxvd2x 46, %[i32],%[ptr_tmp] \n\t" "lxvd2x 47, %[i48],%[ptr_tmp] \n\t" //choose smaller from first and second part "xvcmpgtdp 4, 0,5 \n\t" "xxsel 3, 0,5,4 \n\t" "xxsel 33,32,34,4 \n\t" //load next 64 "lxvd2x 48, %[i64],%[ptr_tmp] \n\t" "lxvd2x 49, %[i80],%[ptr_tmp] \n\t" "vaddudm 1,1,5 \n\t" // get real index for first smaller "lxvd2x 50, %[i96],%[ptr_tmp] \n\t" "lxvd2x 51,%[i112],%[ptr_tmp] \n\t" //compare with previous to get vec_min_index(v6 | vs38 ) and vec_min_value (vs39) "xvcmpgtdp 2,39, 3 \n\t" "xxsel 39,39,3,2 \n\t" "xxsel 38,38,33,2 \n\t" //update index += 8 "vaddudm 5,5,4 \n\t" "xvabsdp 44, 44 \n\t" "xvabsdp 45, 45 \n\t" "xvabsdp 46, 46 \n\t" "xvabsdp 47, 47 \n\t" //update index += 8 "vaddudm 5,5,4 \n\t" "xvabsdp 48, 48 \n\t" "xvabsdp 49, 49 \n\t" "xvabsdp 50, 50 \n\t" "xvabsdp 51, 51 \n\t" //decrement n "addic. %[n], %[n], -32 \n\t" //Loop back if >0 "bgt+ 1b \n\t" //============================================================================== "xvcmpgtdp 2,44,45 \n\t " "xvcmpgtdp 3,46,47 \n\t " "xvcmpgtdp 4,48,49 \n\t " "xvcmpgtdp 5,50,51 \n\t" "xxsel 32,40,41,2 \n\t" "xxsel 0,44,45,2 \n\t" "xxsel 33,42,43,3 \n\t" "xxsel 1,46,47,3 \n\t" "xxsel 34,40,41,4 \n\t" "xxsel 45,48,49,4 \n\t" "xxsel 35,42,43,5 \n\t" "xxsel 47,50,51,5 \n\t" "xvcmpgtdp 2,0, 1 \n\t" "xvcmpgtdp 3, 45,47 \n\t" "xxsel 32,32,33,2 \n\t" "xxsel 0 ,0,1,2 \n\t" "xxsel 34,34,35,3 \n\t" "xxsel 5,45,47,3 \n\t" // for {second 8 elements } we have to add 8 to each so that it became {from 8 to 16} "vaddudm 2,2,4 \n\t" // vs34=vs34 + vs36{8,8} //choose smaller from first and second part "xvcmpgtdp 4, 0,5 \n\t" "xxsel 3, 0,5,4 \n\t" "xxsel 33,32,34,4 \n\t" "vaddudm 1,1,5 \n\t" // get real index for first smaller //compare with previous to get vec_min_index(v6 | vs38 ) and vec_min_value (vs39) "xvcmpgtdp 2,39, 3 \n\t" "xxsel 39,39,3,2 \n\t" "xxsel 38,38,33,2 \n\t" ///////extract min value and min index from vector "xxspltd 32,38,1 \n\t" "xxspltd 40,39,1 \n\t" "xvcmpeqdp. 2, 40,39 \n\t" //cr6 0 bit set if all true, cr6=4*6+bit_ind=24,0011at CR(BI)==1, at=10 hint that it occurs rarely //0b001110=14 "bc 14,24, 3f \n\t" "xvcmpgtdp 4,39, 40 \n\t" "xxsel 0,39,40,4 \n\t" "xxsel 1,38,32,4 \n\t" "stxsdx 0,0,%[ptr_minf] \n\t" "b 4f \n\t" "3: \n\t" //if elements value are equal then choose minimum index "xxspltd 0,40,0 \n\t" "vminud 0,0,6 \n\t" //vs32 vs38 "xxlor 1,32,32 \n\t" "stxsdx 0,0,%[ptr_minf] \n\t" "4: \n\t" "mfvsrd %[index],1 \n\t" : [minf] "=m"(*minf),[ptr_tmp] "+&b"(x),[index] "=r"(index), [n] "+&r"(n) : [mem] "m"(*(const double (*)[n])x), [ptr_x] "b"(x), [ptr_minf] "b"(minf) , [i16] "b"(16), [i32] "b"(32), [i48] "b"(48), [i64] "b"(64), [i80] "b"(80), [i96] "b"(96), [i112] "b"(112), [start] "v"(start), [adder] "v"(temp_add_index) : "cc", "vs0", "vs1","vs2","vs3", "vs4","vs5","vs32", "vs33", "vs34", "vs35", "vs36", "vs37", "vs38", "vs39", "vs40", "vs41", "vs42", "vs43", "vs44", "vs45", "vs46", "vs47", "vs48", "vs49", "vs50", "vs51" ); return index; } BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) { BLASLONG i = 0; BLASLONG j = 0; BLASLONG min = 0; FLOAT minf = 0.0; if (n <= 0 || inc_x <= 0) return (min); minf = ABS(x[0]); //index's not incremented if (inc_x == 1) { BLASLONG n1 = n & -32; if (n1 > 0) { min = diamin_kernel_32(n1, x, &minf); i = n1; } while (i < n) { if (ABS(x[i]) < minf) { min = i; minf = ABS(x[i]); } i++; } return (min + 1); } else { BLASLONG n1 = n & -4; while (j < n1) { if (ABS(x[i]) < minf) { min = j; minf = ABS(x[i]); } if (ABS(x[i + inc_x]) < minf) { min = j + 1; minf = ABS(x[i + inc_x]); } if (ABS(x[i + 2 * inc_x]) < minf) { min = j + 2; minf = ABS(x[i + 2 * inc_x]); } if (ABS(x[i + 3 * inc_x]) < minf) { min = j + 3; minf = ABS(x[i + 3 * inc_x]); } i += inc_x * 4; j += 4; } while (j < n) { if (ABS(x[i]) < minf) { min = j; minf = ABS(x[i]); } i += inc_x; j++; } return (min + 1); } }