/******************************************************************************* Copyright (c) 2015, 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. *******************************************************************************/ #define ASSEMBLER #include "common.h" /* X0 X1 X2 s0 X3 x4 x5 x6 x7*/ /*int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha0,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc, BLASLONG offset) */ #define origM x0 #define origN x1 #define origK x2 #define origPA x3 #define origPB x4 #define pC x5 #define LDC x6 #define offset x7 #define counterL x8 #define counterI x9 #define counterJ x10 #define pB x11 #define pCRow0 x12 #define pCRow1 x13 #define pCRow2 x14 #define pA x15 #define temp x16 #define tempOffset x17 #define tempK x18 #define alpha0 d10 #define alphaV0 v10.d[0] #define alpha1 d11 #define alphaV1 v11.d[0] #define alpha2 d14 #define alphaV2 v14.d[0] #define alpha3 d15 #define alphaV3 v15.d[0] // 00 origM // 01 origN // 02 origK // 03 origPA // 04 origPB // 05 pC // 06 origLDC -> LDC // 07 offset // 08 counterL // 09 counterI // 10 counterJ // 11 pB // 12 pCRow0 // 13 pCRow1 // 14 pCRow2 // 15 pA // 16 temp // 17 tempOffset // 18 must save tempK // 19 must save // 20 must save // 21 must save // 22 must save // 23 must save // 24 must save // 25 must save // 26 must save // 27 must save // 28 must save // 29 frame // 30 link // 31 sp //v00 ALPHA -> pA00, pA01 //v01 pA02, pA03 //v02 //v03 //v04 pA10, pA11 //v05 pA12, pA13 //v06 //v07 //v08 must save pB00, pB01 //v09 must save pB02, pB03 //v10 must save ALPHA0 //v11 must save ALPHA1 //v12 must save pB10, pB11 //v13 must save pB12, pB13 //v14 must save ALPHA2 //v15 must save ALPHA3 //v16 must save C00, C01 //v17 must save C02, C03 //v18 //v19 //v20 C10, C11 //v21 C12, C13 //v22 //v23 //v24 C20, C21 //v25 C22, C23 //v26 //v27 //v28 C30, C31 //v29 C32, C33 //v30 //v31 /******************************************************************************* * Macro definitions *******************************************************************************/ .macro INIT4x4 fmov d16, xzr fmov d17, d16 fmov d20, d17 fmov d21, d16 fmov d24, d17 fmov d25, d16 fmov d28, d17 fmov d29, d16 .endm .macro KERNEL4x4_I ld1 {v8.2d, v9.2d}, [pB] add pB, pB, #32 ld1 {v0.2d, v1.2d}, [pA] add pA, pA, #32 fmul v16.2d, v0.2d, v8.2d[0] fmul v29.2d, v1.2d, v9.2d[1] fmul v20.2d, v0.2d, v8.2d[1] fmul v25.2d, v1.2d, v9.2d[0] fmul v24.2d, v0.2d, v9.2d[0] fmul v21.2d, v1.2d, v8.2d[1] fmul v28.2d, v0.2d, v9.2d[1] fmul v17.2d, v1.2d, v8.2d[0] ld1 {v12.2d, v13.2d}, [pB] add pB, pB, #32 ld1 {v4.2d, v5.2d}, [pA] add pA, pA, #32 .endm .macro KERNEL4x4_M1 fmla v16.2d, v0.2d, v8.2d[0] fmla v29.2d, v1.2d, v9.2d[1] ld1 {v12.2d, v13.2d}, [pB] // For next round add pB, pB, #32 fmla v20.2d, v0.2d, v8.2d[1] fmla v25.2d, v1.2d, v9.2d[0] ld1 {v4.2d, v5.2d}, [pA] // For next round add pA, pA, #32 fmla v24.2d, v0.2d, v9.2d[0] fmla v21.2d, v1.2d, v8.2d[1] prfm PLDL1KEEP, [pA, #512] fmla v28.2d, v0.2d, v9.2d[1] fmla v17.2d, v1.2d, v8.2d[0] .endm .macro KERNEL4x4_M2 fmla v16.2d, v4.2d, v12.2d[0] fmla v29.2d, v5.2d, v13.2d[1] ld1 {v8.2d, v9.2d}, [pB] // For next round add pB, pB, #32 fmla v20.2d, v4.2d, v12.2d[1] fmla v25.2d, v5.2d, v13.2d[0] ld1 {v0.2d, v1.2d}, [pA] // For next round add pA, pA, #32 fmla v24.2d, v4.2d, v13.2d[0] fmla v21.2d, v5.2d, v12.2d[1] prfm PLDL1KEEP, [pB, #512] fmla v28.2d, v4.2d, v13.2d[1] fmla v17.2d, v5.2d, v12.2d[0] .endm .macro KERNEL4x4_E fmla v16.2d, v4.2d, v12.2d[0] fmla v29.2d, v5.2d, v13.2d[1] fmla v20.2d, v4.2d, v12.2d[1] fmla v25.2d, v5.2d, v13.2d[0] fmla v24.2d, v4.2d, v13.2d[0] fmla v21.2d, v5.2d, v12.2d[1] fmla v28.2d, v4.2d, v13.2d[1] fmla v17.2d, v5.2d, v12.2d[0] .endm .macro KERNEL4x4_SUB ld1 {v8.2d, v9.2d}, [pB] add pB, pB, #32 ld1 {v0.2d, v1.2d}, [pA] add pA, pA, #32 fmla v16.2d, v0.2d, v8.2d[0] fmla v29.2d, v1.2d, v9.2d[1] fmla v20.2d, v0.2d, v8.2d[1] fmla v25.2d, v1.2d, v9.2d[0] fmla v24.2d, v0.2d, v9.2d[0] fmla v21.2d, v1.2d, v8.2d[1] fmla v28.2d, v0.2d, v9.2d[1] fmla v17.2d, v1.2d, v8.2d[0] .endm .macro SAVE4x4 fmul v8.2d, v16.2d, alphaV0 fmul v9.2d, v17.2d, alphaV1 st1 {v8.2d, v9.2d}, [pCRow0] add pCRow1, pCRow0, LDC fmul v12.2d, v20.2d, alphaV2 fmul v13.2d, v21.2d, alphaV3 st1 {v12.2d, v13.2d}, [pCRow1] add pCRow2, pCRow1, LDC fmul v8.2d, v24.2d, alphaV0 fmul v9.2d, v25.2d, alphaV1 st1 {v8.2d, v9.2d}, [pCRow2] add pCRow1, pCRow2, LDC fmul v12.2d, v28.2d, alphaV2 fmul v13.2d, v29.2d, alphaV3 st1 {v12.2d, v13.2d}, [pCRow1] add pCRow0, pCRow0, #32 .endm /******************************************************************************/ .macro INIT2x4 fmov d16, xzr fmov d20, d16 fmov d24, d20 fmov d28, d16 .endm .macro KERNEL2x4_SUB ld1 {v8.2d, v9.2d}, [pB] add pB, pB, #32 ld1 {v0.2d}, [pA] add pA, pA, #16 fmla v16.2d, v0.2d, v8.2d[0] fmla v20.2d, v0.2d, v8.2d[1] fmla v24.2d, v0.2d, v9.2d[0] fmla v28.2d, v0.2d, v9.2d[1] .endm .macro SAVE2x4 fmul v8.2d, v16.2d, alphaV0 st1 {v8.2d}, [pCRow0] add pCRow1, pCRow0, LDC fmul v12.2d, v20.2d, alphaV1 st1 {v12.2d}, [pCRow1] add pCRow2, pCRow1, LDC fmul v8.2d, v24.2d, alphaV2 st1 {v8.2d}, [pCRow2] add pCRow1, pCRow2, LDC fmul v12.2d, v28.2d, alphaV3 st1 {v12.2d}, [pCRow1] add pCRow0, pCRow0, #16 .endm /******************************************************************************/ .macro INIT1x4 fmov d16, xzr fmov d20, d16 .endm .macro KERNEL1x4_SUB ldr d0, [pA] add pA, pA, #8 ld1 {v8.2d, v9.2d}, [pB] add pB, pB, #32 fmla v16.2d, v8.2d, v0.d[0] fmla v20.2d, v9.2d, v0.d[0] .endm .macro SAVE1x4 add pCRow1, pCRow0, LDC fmul v8.2d, v16.2d, alphaV0 st1 {v8.d}[0], [pCRow0] st1 {v8.d}[1], [pCRow1] add pCRow2, pCRow1, LDC add pCRow1, pCRow2, LDC fmul v12.2d, v20.2d, alphaV1 st1 {v12.d}[0], [pCRow2] st1 {v12.d}[1], [pCRow1] add pCRow0, pCRow0, #8 .endm /******************************************************************************/ .macro INIT4x2 fmov d16, xzr fmov d17, d16 fmov d20, d17 fmov d21, d16 .endm .macro KERNEL4x2_SUB ld1 {v8.2d}, [pB] add pB, pB, #16 ld1 {v0.2d, v1.2d}, [pA] add pA, pA, #32 fmla v16.2d, v0.2d, v8.2d[0] fmla v17.2d, v1.2d, v8.2d[0] fmla v20.2d, v0.2d, v8.2d[1] fmla v21.2d, v1.2d, v8.2d[1] .endm .macro SAVE4x2 fmul v8.2d, v16.2d, alphaV0 fmul v9.2d, v17.2d, alphaV1 st1 {v8.2d, v9.2d}, [pCRow0] add pCRow1, pCRow0, LDC fmul v12.2d, v20.2d, alphaV2 fmul v13.2d, v21.2d, alphaV3 st1 {v12.2d, v13.2d}, [pCRow1] add pCRow0, pCRow0, #32 .endm /******************************************************************************/ .macro INIT2x2 fmov d16, xzr fmov d20, d16 .endm .macro KERNEL2x2_SUB ld1 {v8.2d}, [pB] add pB, pB, #16 ld1 {v0.2d}, [pA] add pA, pA, #16 fmla v16.2d, v0.2d, v8.2d[0] fmla v20.2d, v0.2d, v8.2d[1] .endm .macro SAVE2x2 fmul v8.2d, v16.2d, alphaV0 st1 {v8.2d}, [pCRow0] add pCRow1 , pCRow0, LDC fmul v12.2d, v20.2d, alphaV1 st1 {v12.2d}, [pCRow1] add pCRow0, pCRow0, #16 .endm /******************************************************************************/ .macro INIT1x2 fmov d16, xzr .endm .macro KERNEL1x2_SUB ld1 {v8.2d} , [pB] add pB , pB, #16 ldr d0 , [pA] add pA, pA, #8 fmla v16.2d, v8.2d, v0.2d[0] .endm .macro SAVE1x2 add pCRow1 , pCRow0, LDC fmul v8.2d, v16.2d, alphaV0 st1 {v8.d}[0], [pCRow0] st1 {v8.d}[1], [pCRow1] add pCRow0, pCRow0, #8 .endm /******************************************************************************/ .macro INIT4x1 fmov d16, xzr fmov d17, d16 .endm .macro KERNEL4x1_SUB ldr d8, [pB] add pB , pB, #8 ld1 {v0.2d, v1.2d}, [pA] add pA , pA, #32 fmla v16.2d, v0.2d, v8.2d[0] fmla v17.2d, v1.2d, v8.2d[0] .endm .macro SAVE4x1 fmul v8.2d, v16.2d, alphaV0 fmul v9.2d, v17.2d, alphaV1 st1 {v8.2d, v9.2d}, [pCRow0] add pCRow0, pCRow0, #32 .endm /******************************************************************************/ .macro INIT2x1 fmov d16, xzr .endm .macro KERNEL2x1_SUB ldr d8, [pB] add pB , pB, #8 ld1 {v0.2d}, [pA] add pA , pA, #16 fmla v16.2d, v0.2d, v8.2d[0] .endm .macro SAVE2x1 fmul v8.2d, v16.2d, alphaV0 st1 {v8.2d}, [pCRow0] add pCRow0, pCRow0, #16 .endm /******************************************************************************/ .macro INIT1x1 fmov d16, xzr .endm .macro KERNEL1x1_SUB ldr d8, [pB] add pB , pB, #8 ldr d0, [pA] add pA , pA, #8 fmadd d16, d0, d8, d16 .endm .macro SAVE1x1 fmul d8, d16, alpha0 str d8, [pCRow0] add pCRow0, pCRow0, #8 .endm /******************************************************************************* * End of macro definitions *******************************************************************************/ PROLOGUE .align 5 add sp, sp, #-(11 * 16) stp d8, d9, [sp, #(0 * 16)] stp d10, d11, [sp, #(1 * 16)] stp d12, d13, [sp, #(2 * 16)] stp d14, d15, [sp, #(3 * 16)] stp d16, d17, [sp, #(4 * 16)] stp x18, x19, [sp, #(5 * 16)] stp x20, x21, [sp, #(6 * 16)] stp x22, x23, [sp, #(7 * 16)] stp x24, x25, [sp, #(8 * 16)] stp x26, x27, [sp, #(9 * 16)] str x28, [sp, #(10 * 16)] fmov alpha0, d0 fmov alpha1, d0 fmov alpha2, d0 fmov alpha3, d0 lsl LDC, LDC, #3 // ldc = ldc * 8 #if !defined(LEFT) neg tempOffset, offset #endif mov pB, origPB mov counterJ, origN asr counterJ, counterJ, #2 // J = J / 4 cmp counterJ, #0 ble dtrmm_kernel_L2_BEGIN /******************************************************************************/ dtrmm_kernel_L4_BEGIN: mov pCRow0, pC // pCRow0 = C add pC, pC, LDC, lsl #2 #if defined(LEFT) mov tempOffset, offset #endif mov pA, origPA // pA = start of A array dtrmm_kernel_L4_M4_BEGIN: mov counterI, origM asr counterI, counterI, #2 // counterI = counterI / 4 cmp counterI, #0 ble dtrmm_kernel_L4_M2_BEGIN dtrmm_kernel_L4_M4_20: #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) mov pB, origPB #else mov pB, origPB lsl temp, tempOffset, #5 add pB, pB, temp add pA, pA, temp #endif #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) sub tempK, origK, tempOffset #elif defined(LEFT) add tempK, tempOffset, #4 #else add tempK, tempOffset, #4 #endif asr counterL , tempK, #1 // L = K / 2 cmp counterL , #2 // is there at least 4 to do? blt dtrmm_kernel_L4_M4_32 KERNEL4x4_I // do one in the K KERNEL4x4_M2 // do another in the K subs counterL, counterL, #2 ble dtrmm_kernel_L4_M4_22a .align 5 dtrmm_kernel_L4_M4_22: KERNEL4x4_M1 KERNEL4x4_M2 subs counterL, counterL, #1 bgt dtrmm_kernel_L4_M4_22 dtrmm_kernel_L4_M4_22a: KERNEL4x4_M1 KERNEL4x4_E b dtrmm_kernel_L4_M4_44 dtrmm_kernel_L4_M4_32: tst counterL, #1 ble dtrmm_kernel_L4_M4_40 KERNEL4x4_I KERNEL4x4_E b dtrmm_kernel_L4_M4_44 dtrmm_kernel_L4_M4_40: INIT4x4 dtrmm_kernel_L4_M4_44: ands counterL , tempK, #1 ble dtrmm_kernel_L4_M4_100 dtrmm_kernel_L4_M4_46: KERNEL4x4_SUB dtrmm_kernel_L4_M4_100: SAVE4x4 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) sub tempK, origK, tempOffset #if defined(LEFT) sub tempK, tempK, #4 #else sub tempK, tempK, #4 #endif lsl temp, tempK, #5 add pA, pA, temp add pB, pB, temp #endif #if defined(LEFT) add tempOffset, tempOffset, #4 #endif dtrmm_kernel_L4_M4_END: subs counterI, counterI, #1 bne dtrmm_kernel_L4_M4_20 dtrmm_kernel_L4_M2_BEGIN: mov counterI, origM tst counterI , #3 ble dtrmm_kernel_L4_END tst counterI, #2 // counterI = counterI / 2 ble dtrmm_kernel_L4_M1_BEGIN dtrmm_kernel_L4_M2_20: INIT2x4 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) mov pB, origPB #else mov pB, origPB lsl temp, tempOffset, #4 add pA, pA, temp lsl temp, tempOffset, #5 add pB, pB, temp #endif #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) sub tempK, origK, tempOffset #elif defined(LEFT) add tempK, tempOffset, #2 #else add tempK, tempOffset, #4 #endif asr counterL , tempK, #3 // counterL = counterL / 8 cmp counterL , #0 ble dtrmm_kernel_L4_M2_40 dtrmm_kernel_L4_M2_22: KERNEL2x4_SUB KERNEL2x4_SUB KERNEL2x4_SUB KERNEL2x4_SUB KERNEL2x4_SUB KERNEL2x4_SUB KERNEL2x4_SUB KERNEL2x4_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L4_M2_22 dtrmm_kernel_L4_M2_40: ands counterL , tempK, #7 // counterL = counterL % 8 ble dtrmm_kernel_L4_M2_100 dtrmm_kernel_L4_M2_42: KERNEL2x4_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L4_M2_42 dtrmm_kernel_L4_M2_100: SAVE2x4 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) sub tempK, origK, tempOffset #if defined(LEFT) sub tempK, tempK, #2 #else sub tempK, tempK, #4 #endif lsl temp, tempK, #4 add pA, pA, temp lsl temp, tempK, #5 add pB, pB, temp #endif #if defined(LEFT) add tempOffset, tempOffset, #2 #endif dtrmm_kernel_L4_M2_END: dtrmm_kernel_L4_M1_BEGIN: tst counterI, #1 // counterI = counterI % 2 ble dtrmm_kernel_L4_END dtrmm_kernel_L4_M1_20: INIT1x4 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) mov pB, origPB #else mov pB, origPB lsl temp, tempOffset, #5 add pB, pB, temp lsl temp, tempOffset, #3 add pA, pA, temp #endif #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) sub tempK, origK, tempOffset #elif defined(LEFT) add tempK, tempOffset, #1 #else add tempK, tempOffset, #4 #endif asr counterL , tempK, #3 // counterL = counterL / 8 cmp counterL , #0 ble dtrmm_kernel_L4_M1_40 dtrmm_kernel_L4_M1_22: KERNEL1x4_SUB KERNEL1x4_SUB KERNEL1x4_SUB KERNEL1x4_SUB KERNEL1x4_SUB KERNEL1x4_SUB KERNEL1x4_SUB KERNEL1x4_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L4_M1_22 dtrmm_kernel_L4_M1_40: ands counterL , tempK, #7 // counterL = counterL % 8 ble dtrmm_kernel_L4_M1_100 dtrmm_kernel_L4_M1_42: KERNEL1x4_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L4_M1_42 dtrmm_kernel_L4_M1_100: SAVE1x4 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) sub tempK, origK, tempOffset #if defined(LEFT) sub tempK, tempK, #1 #else sub tempK, tempK, #4 #endif lsl temp, tempK, #3 add pA, pA, temp lsl temp, tempK, #5 add pB, pB, temp #endif #if defined(LEFT) add tempOffset, tempOffset, #1 #endif dtrmm_kernel_L4_END: lsl temp, origK, #5 add origPB, origPB, temp // B = B + K * 4 * 8 #if !defined(LEFT) add tempOffset, tempOffset, #4 #endif subs counterJ, counterJ , #1 // j-- bgt dtrmm_kernel_L4_BEGIN /******************************************************************************/ dtrmm_kernel_L2_BEGIN: // less than 2 left in N direction mov counterJ , origN tst counterJ , #3 ble dtrmm_kernel_L999 // error, N was less than 4? tst counterJ , #2 ble dtrmm_kernel_L1_BEGIN mov pCRow0, pC // pCRow0 = pC add pC,pC,LDC, lsl #1 #if defined(LEFT) mov tempOffset, offset #endif mov pA, origPA // pA = A dtrmm_kernel_L2_M4_BEGIN: mov counterI, origM asr counterI, counterI, #2 // counterI = counterI / 4 cmp counterI,#0 ble dtrmm_kernel_L2_M2_BEGIN dtrmm_kernel_L2_M4_20: INIT4x2 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) mov pB, origPB #else mov pB, origPB lsl temp, tempOffset, #4 add pB, pB, temp lsl temp, tempOffset, #5 add pA, pA, temp #endif #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) sub tempK, origK, tempOffset #elif defined(LEFT) add tempK, tempOffset, #4 #else add tempK, tempOffset, #2 #endif asr counterL , tempK, #3 // counterL = counterL / 8 cmp counterL,#0 ble dtrmm_kernel_L2_M4_40 .align 5 dtrmm_kernel_L2_M4_22: KERNEL4x2_SUB KERNEL4x2_SUB KERNEL4x2_SUB KERNEL4x2_SUB KERNEL4x2_SUB KERNEL4x2_SUB KERNEL4x2_SUB KERNEL4x2_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L2_M4_22 dtrmm_kernel_L2_M4_40: ands counterL , tempK, #7 // counterL = counterL % 8 ble dtrmm_kernel_L2_M4_100 dtrmm_kernel_L2_M4_42: KERNEL4x2_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L2_M4_42 dtrmm_kernel_L2_M4_100: SAVE4x2 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) sub tempK, origK, tempOffset #if defined(LEFT) sub tempK, tempK, #4 #else sub tempK, tempK, #2 #endif lsl temp, tempK, #5 add pA, pA, temp lsl temp, tempK, #4 add pB, pB, temp #endif #if defined(LEFT) add tempOffset, tempOffset, #4 #endif dtrmm_kernel_L2_M4_END: subs counterI, counterI, #1 bgt dtrmm_kernel_L2_M4_20 dtrmm_kernel_L2_M2_BEGIN: mov counterI, origM tst counterI , #3 ble dtrmm_kernel_L2_END tst counterI, #2 // counterI = counterI / 2 ble dtrmm_kernel_L2_M1_BEGIN dtrmm_kernel_L2_M2_20: INIT2x2 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) mov pB, origPB #else mov pB, origPB lsl temp, tempOffset, #4 add pB, pB, temp lsl temp, tempOffset, #4 add pA, pA, temp #endif #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) sub tempK, origK, tempOffset #elif defined(LEFT) add tempK, tempOffset, #2 #else add tempK, tempOffset, #2 #endif asr counterL , tempK, #3 // counterL = counterL / 8 cmp counterL,#0 ble dtrmm_kernel_L2_M2_40 dtrmm_kernel_L2_M2_22: KERNEL2x2_SUB KERNEL2x2_SUB KERNEL2x2_SUB KERNEL2x2_SUB KERNEL2x2_SUB KERNEL2x2_SUB KERNEL2x2_SUB KERNEL2x2_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L2_M2_22 dtrmm_kernel_L2_M2_40: ands counterL , tempK, #7 // counterL = counterL % 8 ble dtrmm_kernel_L2_M2_100 dtrmm_kernel_L2_M2_42: KERNEL2x2_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L2_M2_42 dtrmm_kernel_L2_M2_100: SAVE2x2 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) sub tempK, origK, tempOffset #if defined(LEFT) sub tempK, tempK, #2 #else sub tempK, tempK, #2 #endif lsl temp, tempK, #4 add pA, pA, temp lsl temp, tempK, #4 add pB, pB, temp #endif #if defined(LEFT) add tempOffset, tempOffset, #2 #endif dtrmm_kernel_L2_M2_END: dtrmm_kernel_L2_M1_BEGIN: tst counterI, #1 // counterI = counterI % 2 ble dtrmm_kernel_L2_END dtrmm_kernel_L2_M1_20: INIT1x2 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) mov pB, origPB #else mov pB, origPB lsl temp, tempOffset, #4 add pB, pB, temp lsl temp, tempOffset, #3 add pA, pA, temp #endif #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) sub tempK, origK, tempOffset #elif defined(LEFT) add tempK, tempOffset, #1 #else add tempK, tempOffset, #2 #endif asr counterL , tempK, #3 // counterL = counterL / 8 cmp counterL, #0 ble dtrmm_kernel_L2_M1_40 dtrmm_kernel_L2_M1_22: KERNEL1x2_SUB KERNEL1x2_SUB KERNEL1x2_SUB KERNEL1x2_SUB KERNEL1x2_SUB KERNEL1x2_SUB KERNEL1x2_SUB KERNEL1x2_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L2_M1_22 dtrmm_kernel_L2_M1_40: ands counterL , tempK, #7 // counterL = counterL % 8 ble dtrmm_kernel_L2_M1_100 dtrmm_kernel_L2_M1_42: KERNEL1x2_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L2_M1_42 dtrmm_kernel_L2_M1_100: SAVE1x2 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) sub tempK, origK, tempOffset #if defined(LEFT) sub tempK, tempK, #1 #else sub tempK, tempK, #2 #endif lsl temp, tempK, #3 add pA, pA, temp lsl temp, tempK, #4 add pB, pB, temp #endif #if defined(LEFT) add tempOffset, tempOffset, #1 #endif dtrmm_kernel_L2_END: #if !defined(LEFT) add tempOffset, tempOffset, #2 #endif add origPB, origPB, origK, lsl #4 // B = B + K * 2 * 8 /******************************************************************************/ dtrmm_kernel_L1_BEGIN: mov counterJ , origN tst counterJ , #1 ble dtrmm_kernel_L999 // done mov pCRow0, pC // pCRow0 = C add pC , pC , LDC // Update pC to point to next #if defined(LEFT) mov tempOffset, offset #endif mov pA, origPA // pA = A dtrmm_kernel_L1_M4_BEGIN: mov counterI, origM asr counterI, counterI, #2 // counterI = counterI / 4 cmp counterI, #0 ble dtrmm_kernel_L1_M2_BEGIN dtrmm_kernel_L1_M4_20: INIT4x1 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) mov pB, origPB #else mov pB, origPB lsl temp, tempOffset, #3 add pB, pB, temp lsl temp, tempOffset, #5 add pA, pA, temp #endif #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) sub tempK, origK, tempOffset #elif defined(LEFT) add tempK, tempOffset, #4 #else add tempK, tempOffset, #1 #endif asr counterL , tempK, #3 // counterL = counterL / 8 cmp counterL , #0 ble dtrmm_kernel_L1_M4_40 .align 5 dtrmm_kernel_L1_M4_22: KERNEL4x1_SUB KERNEL4x1_SUB KERNEL4x1_SUB KERNEL4x1_SUB KERNEL4x1_SUB KERNEL4x1_SUB KERNEL4x1_SUB KERNEL4x1_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L1_M4_22 dtrmm_kernel_L1_M4_40: ands counterL , tempK, #7 // counterL = counterL % 8 ble dtrmm_kernel_L1_M4_100 dtrmm_kernel_L1_M4_42: KERNEL4x1_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L1_M4_42 dtrmm_kernel_L1_M4_100: SAVE4x1 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) sub tempK, origK, tempOffset #if defined(LEFT) sub tempK, tempK, #4 #else sub tempK, tempK, #1 #endif lsl temp, tempK, #5 add pA, pA, temp lsl temp, tempK, #3 add pB, pB, temp #endif #if defined(LEFT) add tempOffset, tempOffset, #4 #endif dtrmm_kernel_L1_M4_END: subs counterI, counterI, #1 bgt dtrmm_kernel_L1_M4_20 dtrmm_kernel_L1_M2_BEGIN: mov counterI, origM tst counterI , #3 ble dtrmm_kernel_L1_END tst counterI, #2 // counterI = counterI / 2 ble dtrmm_kernel_L1_M1_BEGIN dtrmm_kernel_L1_M2_20: INIT2x1 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) mov pB, origPB #else mov pB, origPB lsl temp, tempOffset, #3 add pB, pB, temp lsl temp, tempOffset, #4 add pA, pA, temp #endif #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) sub tempK, origK, tempOffset #elif defined(LEFT) add tempK, tempOffset, #2 #else add tempK, tempOffset, #1 #endif asr counterL , tempK, #3 // counterL = counterL / 8 cmp counterL , #0 ble dtrmm_kernel_L1_M2_40 dtrmm_kernel_L1_M2_22: KERNEL2x1_SUB KERNEL2x1_SUB KERNEL2x1_SUB KERNEL2x1_SUB KERNEL2x1_SUB KERNEL2x1_SUB KERNEL2x1_SUB KERNEL2x1_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L1_M2_22 dtrmm_kernel_L1_M2_40: ands counterL , tempK, #7 // counterL = counterL % 8 ble dtrmm_kernel_L1_M2_100 dtrmm_kernel_L1_M2_42: KERNEL2x1_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L1_M2_42 dtrmm_kernel_L1_M2_100: SAVE2x1 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) sub tempK, origK, tempOffset #if defined(LEFT) sub tempK, tempK, #2 #else sub tempK, tempK, #1 #endif lsl temp, tempK, #4 add pA, pA, temp lsl temp, tempK, #3 add pB, pB, temp #endif #if defined(LEFT) add tempOffset, tempOffset, #2 #endif dtrmm_kernel_L1_M2_END: dtrmm_kernel_L1_M1_BEGIN: tst counterI, #1 // counterI = counterI % 2 ble dtrmm_kernel_L1_END dtrmm_kernel_L1_M1_20: INIT1x1 #if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) mov pB, origPB #else mov pB, origPB lsl temp, tempOffset, #3 add pB, pB, temp lsl temp, tempOffset, #3 add pA, pA, temp #endif #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) sub tempK, origK, tempOffset #elif defined(LEFT) add tempK, tempOffset, #1 #else add tempK, tempOffset, #1 #endif asr counterL , tempK, #3 // counterL = counterL / 8 cmp counterL , #0 ble dtrmm_kernel_L1_M1_40 dtrmm_kernel_L1_M1_22: KERNEL1x1_SUB KERNEL1x1_SUB KERNEL1x1_SUB KERNEL1x1_SUB KERNEL1x1_SUB KERNEL1x1_SUB KERNEL1x1_SUB KERNEL1x1_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L1_M1_22 dtrmm_kernel_L1_M1_40: ands counterL , tempK, #7 // counterL = counterL % 8 ble dtrmm_kernel_L1_M1_100 dtrmm_kernel_L1_M1_42: KERNEL1x1_SUB subs counterL, counterL, #1 bgt dtrmm_kernel_L1_M1_42 dtrmm_kernel_L1_M1_100: SAVE1x1 dtrmm_kernel_L1_END: dtrmm_kernel_L999: mov x0, #0 // set return value ldp d8, d9, [sp, #(0 * 16)] ldp d10, d11, [sp, #(1 * 16)] ldp d12, d13, [sp, #(2 * 16)] ldp d14, d15, [sp, #(3 * 16)] ldp d16, d17, [sp, #(4 * 16)] ldp x18, x19, [sp, #(5 * 16)] ldp x20, x21, [sp, #(6 * 16)] ldp x22, x23, [sp, #(7 * 16)] ldp x24, x25, [sp, #(8 * 16)] ldp x26, x27, [sp, #(9 * 16)] ldr x28, [sp, #(10 * 16)] add sp, sp, #(11*16) ret EPILOGUE