/*************************************************************************** Copyright (c) 2013, 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. *****************************************************************************/ /************************************************************************************** * 2013/11/29 Saar * BLASTEST : OK * CTEST : OK * TEST : OK * **************************************************************************************/ #define ASSEMBLER #include "common.h" #define STACKSIZE 256 #define OLD_LDA [fp, #0 ] #define X [fp, #4 ] #define OLD_INC_X [fp, #8 ] #define Y [fp, #12 ] #define OLD_INC_Y [fp, #16 ] #define OLD_A r3 #define OLD_M r0 #define AO1 r0 #define N r1 #define J r2 #define AO2 r4 #define XO r5 #define YO r6 #define LDA r7 #define INC_X r8 #define INC_Y r9 #define I r12 #define FP_ZERO [fp, #-228] #define FP_ZERO_0 [fp, #-228] #define FP_ZERO_1 [fp, #-224] #define ALPHA_I [fp, #-236] #define ALPHA_R [fp, #-244] #define M [fp, #-252 ] #define A [fp, #-256 ] #define X_PRE 64 #define Y_PRE 0 #define A_PRE 0 /**************************************************************************************/ #if !defined(CONJ) && !defined(XCONJ) #define KMAC_R fnmacs #define KMAC_I fmacs #define FMAC_R1 fmacs #define FMAC_R2 fnmacs #define FMAC_I1 fmacs #define FMAC_I2 fmacs #elif defined(CONJ) && !defined(XCONJ) #define KMAC_R fmacs #define KMAC_I fnmacs #define FMAC_R1 fmacs #define FMAC_R2 fnmacs #define FMAC_I1 fmacs #define FMAC_I2 fmacs #elif !defined(CONJ) && defined(XCONJ) #define KMAC_R fmacs #define KMAC_I fnmacs #define FMAC_R1 fmacs #define FMAC_R2 fmacs #define FMAC_I1 fnmacs #define FMAC_I2 fmacs #else #define KMAC_R fnmacs #define KMAC_I fmacs #define FMAC_R1 fmacs #define FMAC_R2 fmacs #define FMAC_I1 fnmacs #define FMAC_I2 fmacs #endif .macro INIT_F4 pld [ YO, #Y_PRE ] flds s8 , FP_ZERO vmov.f32 s9 , s8 vmov.f32 s10, s8 vmov.f32 s11, s8 vmov.f32 s12, s8 vmov.f32 s13, s8 vmov.f32 s14, s8 vmov.f32 s15, s8 .endm .macro KERNEL_F4X4 pld [ XO, #X_PRE ] KERNEL_F4X1 KERNEL_F4X1 KERNEL_F4X1 KERNEL_F4X1 .endm .macro KERNEL_F4X1 pld [ AO2, #A_PRE ] flds s0 , [ AO1 ] flds s1 , [ AO1, #4 ] flds s2 , [ AO1, #8 ] flds s3 , [ AO1, #12 ] flds s4 , [ XO ] flds s5 , [ XO, #4 ] fmacs s8 , s0, s4 fmacs s9 , s0, s5 fmacs s10 , s2, s4 fmacs s11 , s2, s5 KMAC_R s8 , s1, s5 KMAC_I s9 , s1, s4 KMAC_R s10 , s3, s5 KMAC_I s11 , s3, s4 flds s0 , [ AO1, #16 ] flds s1 , [ AO1, #20 ] flds s2 , [ AO1, #24 ] flds s3 , [ AO1, #28 ] fmacs s12 , s0, s4 fmacs s13 , s0, s5 fmacs s14 , s2, s4 fmacs s15 , s2, s5 KMAC_R s12 , s1, s5 KMAC_I s13 , s1, s4 KMAC_R s14 , s3, s5 KMAC_I s15 , s3, s4 add XO , XO, #8 add AO1 , AO1, LDA add AO2 , AO2, LDA .endm .macro SAVE_F4 flds s0, ALPHA_R flds s1, ALPHA_I fldmias YO, { s4 - s7 } FMAC_R1 s4 , s0 , s8 FMAC_I1 s5 , s0 , s9 FMAC_R2 s4 , s1 , s9 FMAC_I2 s5 , s1 , s8 FMAC_R1 s6 , s0 , s10 FMAC_I1 s7 , s0 , s11 FMAC_R2 s6 , s1 , s11 FMAC_I2 s7 , s1 , s10 fstmias YO!, { s4 - s7 } fldmias YO, { s4 - s7 } FMAC_R1 s4 , s0 , s12 FMAC_I1 s5 , s0 , s13 FMAC_R2 s4 , s1 , s13 FMAC_I2 s5 , s1 , s12 FMAC_R1 s6 , s0 , s14 FMAC_I1 s7 , s0 , s15 FMAC_R2 s6 , s1 , s15 FMAC_I2 s7 , s1 , s14 fstmias YO!, { s4 - s7 } .endm .macro INIT_F1 flds s8 , FP_ZERO vmov.f32 s9 , s8 .endm .macro KERNEL_F1X1 flds s0 , [ AO1 ] flds s1 , [ AO1, #4 ] flds s4 , [ XO ] flds s5 , [ XO, #4 ] fmacs s8 , s0, s4 fmacs s9 , s0, s5 KMAC_R s8 , s1, s5 KMAC_I s9 , s1, s4 add XO , XO, #8 add AO1 , AO1, LDA .endm .macro SAVE_F1 flds s0, ALPHA_R flds s1, ALPHA_I fldmias YO, { s4 - s5 } FMAC_R1 s4 , s0 , s8 FMAC_I1 s5 , s0 , s9 FMAC_R2 s4 , s1 , s9 FMAC_I2 s5 , s1 , s8 fstmias YO, { s4 - s5 } add YO, YO, #8 .endm /****************************************************************************************/ .macro INIT_S4 flds s8 , FP_ZERO vmov.f32 s9 , s8 vmov.f32 s10, s8 vmov.f32 s11, s8 vmov.f32 s12, s8 vmov.f32 s13, s8 vmov.f32 s14, s8 vmov.f32 s15, s8 .endm .macro KERNEL_S4X4 KERNEL_S4X1 KERNEL_S4X1 KERNEL_S4X1 KERNEL_S4X1 .endm .macro KERNEL_S4X1 flds s0 , [ AO1 ] flds s1 , [ AO1, #4 ] flds s2 , [ AO1, #8 ] flds s3 , [ AO1, #12 ] flds s4 , [ XO ] flds s5 , [ XO, #4 ] fmacs s8 , s0, s4 fmacs s9 , s0, s5 fmacs s10 , s2, s4 fmacs s11 , s2, s5 KMAC_R s8 , s1, s5 KMAC_I s9 , s1, s4 KMAC_R s10 , s3, s5 KMAC_I s11 , s3, s4 flds s0 , [ AO1, #16 ] flds s1 , [ AO1, #20 ] flds s2 , [ AO1, #24 ] flds s3 , [ AO1, #28 ] fmacs s12 , s0, s4 fmacs s13 , s0, s5 fmacs s14 , s2, s4 fmacs s15 , s2, s5 KMAC_R s12 , s1, s5 KMAC_I s13 , s1, s4 KMAC_R s14 , s3, s5 KMAC_I s15 , s3, s4 add XO , XO, INC_X add AO1 , AO1, LDA add AO2 , AO2, LDA .endm .macro SAVE_S4 flds s0, ALPHA_R flds s1, ALPHA_I fldmias YO, { s4 - s5 } FMAC_R1 s4 , s0 , s8 FMAC_I1 s5 , s0 , s9 FMAC_R2 s4 , s1 , s9 FMAC_I2 s5 , s1 , s8 fstmias YO, { s4 - s5 } add YO, YO, INC_Y fldmias YO, { s6 - s7 } FMAC_R1 s6 , s0 , s10 FMAC_I1 s7 , s0 , s11 FMAC_R2 s6 , s1 , s11 FMAC_I2 s7 , s1 , s10 fstmias YO, { s6 - s7 } add YO, YO, INC_Y fldmias YO, { s4 - s5 } FMAC_R1 s4 , s0 , s12 FMAC_I1 s5 , s0 , s13 FMAC_R2 s4 , s1 , s13 FMAC_I2 s5 , s1 , s12 fstmias YO, { s4 - s5 } add YO, YO, INC_Y fldmias YO, { s6 - s7 } FMAC_R1 s6 , s0 , s14 FMAC_I1 s7 , s0 , s15 FMAC_R2 s6 , s1 , s15 FMAC_I2 s7 , s1 , s14 fstmias YO, { s6 - s7 } add YO, YO, INC_Y .endm .macro INIT_S1 flds s8 , FP_ZERO vmov.f32 s9 , s8 .endm .macro KERNEL_S1X1 flds s0 , [ AO1 ] flds s1 , [ AO1, #4 ] flds s4 , [ XO ] flds s5 , [ XO, #4 ] fmacs s8 , s0, s4 fmacs s9 , s0, s5 KMAC_R s8 , s1, s5 KMAC_I s9 , s1, s4 add XO , XO, INC_X add AO1 , AO1, LDA .endm .macro SAVE_S1 flds s0, ALPHA_R flds s1, ALPHA_I fldmias YO, { s4 - s5 } FMAC_R1 s4 , s0 , s8 FMAC_I1 s5 , s0 , s9 FMAC_R2 s4 , s1 , s9 FMAC_I2 s5 , s1 , s8 fstmias YO, { s4 - s5 } add YO, YO, INC_Y .endm /************************************************************************************** * End of macro definitions **************************************************************************************/ PROLOGUE .align 5 push {r4 - r9 , fp} add fp, sp, #28 sub sp, sp, #STACKSIZE // reserve stack sub r12, fp, #192 #if defined(DOUBLE) vstm r12, { d8 - d15 } // store floating point registers #else vstm r12, { s8 - s15 } // store floating point registers #endif movs r12, #0 str r12, FP_ZERO str r12, FP_ZERO_1 cmp OLD_M, #0 ble cgemvn_kernel_L999 cmp N, #0 ble cgemvn_kernel_L999 str OLD_A, A str OLD_M, M vstr s0 , ALPHA_R vstr s1 , ALPHA_I ldr INC_X , OLD_INC_X ldr INC_Y , OLD_INC_Y cmp INC_X, #0 beq cgemvn_kernel_L999 cmp INC_Y, #0 beq cgemvn_kernel_L999 ldr LDA, OLD_LDA #if defined(DOUBLE) lsl LDA, LDA, #4 // LDA * SIZE * 2 #else lsl LDA, LDA, #3 // LDA * SIZE * 2 #endif cmp INC_X, #1 bne cgemvn_kernel_S4_BEGIN cmp INC_Y, #1 bne cgemvn_kernel_S4_BEGIN cgemvn_kernel_F4_BEGIN: ldr YO , Y ldr I, M asrs I, I, #2 // I = M / 4 ble cgemvn_kernel_F1_BEGIN cgemvn_kernel_F4X4: ldr AO1, A add AO2, AO1, LDA add r3 , AO1, #32 str r3 , A add AO2, AO2, LDA add AO2, AO2, LDA ldr XO , X INIT_F4 asrs J, N, #2 // J = N / 4 ble cgemvn_kernel_F4X1 cgemvn_kernel_F4X4_10: KERNEL_F4X4 subs J, J, #1 bne cgemvn_kernel_F4X4_10 cgemvn_kernel_F4X1: ands J, N , #3 ble cgemvn_kernel_F4_END cgemvn_kernel_F4X1_10: KERNEL_F4X1 subs J, J, #1 bne cgemvn_kernel_F4X1_10 cgemvn_kernel_F4_END: SAVE_F4 subs I , I , #1 bne cgemvn_kernel_F4X4 cgemvn_kernel_F1_BEGIN: ldr I, M ands I, I , #3 ble cgemvn_kernel_L999 cgemvn_kernel_F1X1: ldr AO1, A add r3, AO1, #8 str r3, A ldr XO , X INIT_F1 mov J, N cgemvn_kernel_F1X1_10: KERNEL_F1X1 subs J, J, #1 bne cgemvn_kernel_F1X1_10 cgemvn_kernel_F1_END: SAVE_F1 subs I , I , #1 bne cgemvn_kernel_F1X1 b cgemvn_kernel_L999 /*************************************************************************************************************/ cgemvn_kernel_S4_BEGIN: #if defined(DOUBLE) lsl INC_X, INC_X, #4 // INC_X * SIZE * 2 lsl INC_Y, INC_Y, #4 // INC_Y * SIZE * 2 #else lsl INC_X, INC_X, #3 // INC_X * SIZE * 2 lsl INC_Y, INC_Y, #3 // INC_Y * SIZE * 2 #endif ldr YO , Y ldr I, M asrs I, I, #2 // I = M / 4 ble cgemvn_kernel_S1_BEGIN cgemvn_kernel_S4X4: ldr AO1, A add AO2, AO1, LDA add r3 , AO1, #32 str r3 , A ldr XO , X INIT_S4 asrs J, N, #2 // J = N / 4 ble cgemvn_kernel_S4X1 cgemvn_kernel_S4X4_10: KERNEL_S4X4 subs J, J, #1 bne cgemvn_kernel_S4X4_10 cgemvn_kernel_S4X1: ands J, N , #3 ble cgemvn_kernel_S4_END cgemvn_kernel_S4X1_10: KERNEL_S4X1 subs J, J, #1 bne cgemvn_kernel_S4X1_10 cgemvn_kernel_S4_END: SAVE_S4 subs I , I , #1 bne cgemvn_kernel_S4X4 cgemvn_kernel_S1_BEGIN: ldr I, M ands I, I , #3 ble cgemvn_kernel_L999 cgemvn_kernel_S1X1: ldr AO1, A add r3, AO1, #8 str r3, A ldr XO , X INIT_S1 mov J, N cgemvn_kernel_S1X1_10: KERNEL_S1X1 subs J, J, #1 bne cgemvn_kernel_S1X1_10 cgemvn_kernel_S1_END: SAVE_S1 subs I , I , #1 bne cgemvn_kernel_S1X1 /*************************************************************************************************************/ cgemvn_kernel_L999: sub r3, fp, #192 #if defined(DOUBLE) vldm r3, { d8 - d15 } // restore floating point registers #else vldm r3, { s8 - s15 } // restore floating point registers #endif mov r0, #0 // set return value sub sp, fp, #28 pop {r4 -r9 ,fp} bx lr EPILOGUE