/* BLIS An object-based framework for developing high-performance BLAS-like libraries. Copyright (C) 2014, The University of Texas at Austin Copyright (C) 2021, The University of Tokyo 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" #include "assert.h" // Label locality & misc. #include "../armv8a_asm_utils.h" // Nanokernel operations. #include "../armv8a_asm_d2x2.h" /* Odd-NR dgemmsup_rv_*m kernels are special in that * despite of the row-major name, C is laid out in COLUMNS in the register space. * * Block order: * * +---+ +---+ * | 0 | | 3 | |6 * +---+ +---+ | * +---+ +---+ * | 1 | | 4 | |7 * +---+ +---+ | * +---+ +---+ * | 2 | | 5 | |8 * +---+ +---+ | * */ #define DGEMM_C6X5_MKER_LOOP_PLAIN(C00,C01,C02,C03,C04,C10,C11,C12,C13,C14,C20,C21,C22,C23,C24,A0,A1,A2,B0,B1,B2,BIDX,AELEMADDR,AELEMST,BADDR,BSHIFT,LOADNEXT) \ DGEMM_2X2_NANOKERNEL(C00,C01,A0,B0) \ DGEMM_2X2_NANOKERNEL(C10,C11,A1,B0) \ DGEMM_2X2_NANOKERNEL(C20,C21,A2,B0) \ DGEMM_LOAD1V_ ##LOADNEXT (B0,BADDR,BSHIFT) \ DGEMM_2X2_NANOKERNEL(C02,C03,A0,B1) \ DGEMM_2X2_NANOKERNEL(C12,C13,A1,B1) \ DGEMM_2X2_NANOKERNEL(C22,C23,A2,B1) \ DGEMM_LOAD1V_ ##LOADNEXT (B1,BADDR,BSHIFT+16) \ " fmla v"#C04".2d, v"#A0".2d, v"#B2".d["#BIDX"] \n\t" \ DGEMM_LOAD1V_G_ ##LOADNEXT (A0,AELEMADDR,AELEMST) \ " fmla v"#C14".2d, v"#A1".2d, v"#B2".d["#BIDX"] \n\t" \ DGEMM_LOAD1V_G_ ##LOADNEXT (A1,AELEMADDR,AELEMST) \ " fmla v"#C24".2d, v"#A2".2d, v"#B2".d["#BIDX"] \n\t" // Interleaving load or not. #define DGEMM_LOAD1V_noload(V1,ADDR,IMM) #define DGEMM_LOAD1V_load(V1,ADDR,IMM) \ " ldr q"#V1", ["#ADDR", #"#IMM"] \n\t" #define DGEMM_LOAD1V_G_noload(V1,ADDR,ST) #define DGEMM_LOAD1V_G_load(V1,ADDR,ST) \ " ld1 {v"#V1".d}[0], ["#ADDR"], "#ST" \n\t" \ " ld1 {v"#V1".d}[1], ["#ADDR"], "#ST" \n\t" // Prefetch C in the long direction. #define DPRFMC_FWD(CADDR,DLONGC) \ " prfm PLDL1KEEP, ["#CADDR"] \n\t" \ " add "#CADDR", "#CADDR", "#DLONGC" \n\t" // For column-storage of C. #define DLOADC_3V_C_FWD(C0,C1,C2,CADDR,CSHIFT,CSC) \ DLOAD2V(C0,C1,CADDR,CSHIFT) \ DLOAD1V(C2,CADDR,CSHIFT+32) \ " add "#CADDR", "#CADDR", "#CSC" \n\t" #define DSTOREC_3V_C_FWD(C0,C1,C2,CADDR,CSHIFT,CSC) \ DSTORE2V(C0,C1,CADDR,CSHIFT) \ DSTORE1V(C2,CADDR,CSHIFT+32) \ " add "#CADDR", "#CADDR", "#CSC" \n\t" // For row-storage of C: Store 2+1/2 vectors. #define DLOADC_2PHV_R_FWD(C0,C1,C2,CIDX,CADDR,CSHIFT,RSC,CTMP) \ " add "#CTMP", "#CADDR", "#CSHIFT"+32 \n\t" \ DLOAD2V(C0,C1,CADDR,CSHIFT) \ " ld1 {v"#C2".d}["#CIDX"], ["#CTMP"] \n\t" \ " add "#CADDR", "#CADDR", "#RSC" \n\t" #define DSTOREC_2PHV_R_FWD(C0,C1,C2,CIDX,CADDR,CSHIFT,RSC,CTMP) \ " add "#CTMP", "#CADDR", "#CSHIFT"+32 \n\t" \ DSTORE2V(C0,C1,CADDR,CSHIFT) \ " st1 {v"#C2".d}["#CIDX"], ["#CTMP"] \n\t" \ " add "#CADDR", "#CADDR", "#RSC" \n\t" #define DSCALE5V(V0,V1,V2,V3,V4,A,IDX) \ DSCALE4V(V0,V1,V2,V3,A,IDX) \ DSCALE1V(V4,A,IDX) #define DSCALEA5V(D0,D1,D2,D3,D4,S0,S1,S2,S3,S4,A,IDX) \ DSCALEA4V(D0,D1,D2,D3,S0,S1,S2,S3,A,IDX) \ DSCALEA1V(D4,S4,A,IDX) void bli_dgemmsup_rv_armv8a_asm_6x5m ( conj_t conja, conj_t conjb, dim_t m0, dim_t n0, dim_t k0, const void* alpha, const void* a, inc_t rs_a0, inc_t cs_a0, const void* b, inc_t rs_b0, inc_t cs_b0, const void* beta, void* c, inc_t rs_c0, inc_t cs_c0, auxinfo_t* data, const cntx_t* cntx ) { assert( n0 == 5 ); // LLVM has very bad routing ability for inline asm. // Limit number of registers in case of Clang compilation. #ifndef __clang__ const void* a_next = bli_auxinfo_next_a( data ); const void* b_next = bli_auxinfo_next_b( data ); #endif uint64_t ps_a = bli_auxinfo_ps_a( data ); // Typecast local copies of integers in case dim_t and inc_t are a // different size than is expected by load instructions. uint64_t k_mker = k0 / 6; uint64_t k_left = k0 % 6; int64_t m_iter = m0 / 6; int64_t m_left = m0 % 6; uint64_t rs_a = rs_a0; uint64_t cs_a = cs_a0; uint64_t rs_b = rs_b0; uint64_t rs_c = rs_c0; uint64_t cs_c = cs_c0; // uint64_t cs_b = cs_b0; assert( cs_b0 == 1 ); if ( m_iter == 0 ) goto consider_edge_cases; __asm__ volatile ( " ldr x10, %[a] \n\t" " ldr x13, %[c] \n\t" " ldr x12, %[m_iter] \n\t" " ldr x11, %[ps_a] \n\t" // Panel-skip of A. " ldr x9, %[rs_a] \n\t" // Row-skip of A. " ldr x2, %[cs_a] \n\t" // Column-skip of A. " ldr x3, %[rs_b] \n\t" // Row-skip of B. " \n\t" " ldr x6, %[rs_c] \n\t" // Row-skip of C. " ldr x7, %[cs_c] \n\t" // Column-skip of C. " \n\t" " \n\t" // Multiply some address skips by sizeof(double). " lsl x11, x11, #3 \n\t" // ps_a " lsl x9, x9, #3 \n\t" // rs_a " lsl x2, x2, #3 \n\t" // cs_a " lsl x3, x3, #3 \n\t" // rs_b " lsl x6, x6, #3 \n\t" // rs_c " lsl x7, x7, #3 \n\t" // cs_c " \n\t" " mov x1, x5 \n\t" " cmp x7, #8 \n\t" // Prefetch column-strided C. BNE(C_PREFETCH_COLS) // This prefetch will not cover further mker perts. Skip. // // DPRFMC_FWD(x1,x6) // DPRFMC_FWD(x1,x6) // DPRFMC_FWD(x1,x6) // DPRFMC_FWD(x1,x6) // DPRFMC_FWD(x1,x6) // DPRFMC_FWD(x1,x6) BRANCH(C_PREFETCH_END) LABEL(C_PREFETCH_COLS) DPRFMC_FWD(x1,x7) DPRFMC_FWD(x1,x7) DPRFMC_FWD(x1,x7) DPRFMC_FWD(x1,x7) DPRFMC_FWD(x1,x7) LABEL(C_PREFETCH_END) // // Millikernel. LABEL(MILLIKER_MLOOP) " \n\t" " mov x0, x10 \n\t" // Parameters to be reloaded " mov x5, x13 \n\t" // within each millikernel loop. " ldr x1, %[b] \n\t" " ldr x4, %[k_mker] \n\t" " ldr x8, %[k_left] \n\t" " \n\t" // Storage scheme: // V[ 0:14] <- C // V[15:23] <- A // V[24:29] <- B // Under this scheme, the following is defined: #define DGEMM_C6X5_MKER_LOOP_PLAIN_LOC(A0,A1,A2,B0,B1,B2,BIDX,AELEMADDR,AELEMST,BADDR,BSHIFT,LOADNEXT) \ DGEMM_C6X5_MKER_LOOP_PLAIN(0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,A0,A1,A2,B0,B1,B2,BIDX,AELEMADDR,AELEMST,BADDR,BSHIFT,LOADNEXT) // Load from memory. LABEL(LOAD_ABC) " \n\t" // No-microkernel early return is a must " cmp x4, #0 \n\t" // to avoid out-of-boundary read. BEQ(CLEAR_CCOLS) " \n\t" " mov x14, x0 \n\t" // Load A. " ld1 {v15.d}[0], [x14], x9 \n\t" " ld1 {v15.d}[1], [x14], x9 \n\t" " ld1 {v16.d}[0], [x14], x9 \n\t" " ld1 {v16.d}[1], [x14], x9 \n\t" " ld1 {v17.d}[0], [x14], x9 \n\t" " ld1 {v17.d}[1], [x14], x9 \n\t" " add x0, x0, x2 \n\t" " mov x14, x0 \n\t" " ld1 {v18.d}[0], [x14], x9 \n\t" " ld1 {v18.d}[1], [x14], x9 \n\t" " ld1 {v19.d}[0], [x14], x9 \n\t" " ld1 {v19.d}[1], [x14], x9 \n\t" " ld1 {v20.d}[0], [x14], x9 \n\t" " ld1 {v20.d}[1], [x14], x9 \n\t" " add x0, x0, x2 \n\t" " mov x14, x0 \n\t" " ld1 {v21.d}[0], [x14], x9 \n\t" " ld1 {v21.d}[1], [x14], x9 \n\t" " ld1 {v22.d}[0], [x14], x9 \n\t" " ld1 {v22.d}[1], [x14], x9 \n\t" " ld1 {v23.d}[0], [x14], x9 \n\t" " ld1 {v23.d}[1], [x14], x9 \n\t" " add x0, x0, x2 \n\t" " mov x14, x0 \n\t" " \n\t" " ldr q24, [x1, #16*0] \n\t" // Load B. " ldr q25, [x1, #16*1] \n\t" " ldr d26, [x1, #16*2] \n\t" // Scalar loads into idx 0. " add x1, x1, x3 \n\t" " ldr q27, [x1, #16*0] \n\t" " ldr q28, [x1, #16*1] \n\t" " ldr d29, [x1, #16*2] \n\t" " add x1, x1, x3 \n\t" LABEL(CLEAR_CCOLS) CLEAR4V(0,1,2,3) CLEAR1V(4) CLEAR4V(5,6,7,8) CLEAR1V(9) CLEAR4V(10,11,12,13) CLEAR1V(14) // No-microkernel early return, once again. BEQ(K_LEFT_LOOP) // // Microkernel is defined here as: #define DGEMM_C6X5_MKER_LOOP_PLAIN_LOC_FWD(A0,A1,A2,B0,B1,B2,BIDX) \ DGEMM_C6X5_MKER_LOOP_PLAIN_LOC(A0,A1,A2,B0,B1,B2,BIDX,x14,x9,x1,0,load) \ "ld1 {v"#A2".d}[0], [x14], x9 \n\t" \ "ld1 {v"#A2".d}[1], [x14], x9 \n\t" \ "add x0, x0, x2 \n\t" \ "mov x14, x0 \n\t" \ /* Due to this loading, BIDX can only be 0 here. */ \ "ldr d"#B2", [x1, #16*2] \n\t" \ "add x1, x1, x3 \n\t" // Start microkernel loop. LABEL(K_MKER_LOOP) DGEMM_C6X5_MKER_LOOP_PLAIN_LOC_FWD(15,16,17,24,25,26,0) DGEMM_C6X5_MKER_LOOP_PLAIN_LOC_FWD(18,19,20,27,28,29,0) DGEMM_C6X5_MKER_LOOP_PLAIN_LOC_FWD(21,22,23,24,25,26,0) " \n\t" // Decrease counter before final replica. " subs x4, x4, #1 \n\t" // Branch early to avoid reading excess mem. BEQ(FIN_MKER_LOOP) DGEMM_C6X5_MKER_LOOP_PLAIN_LOC_FWD(15,16,17,27,28,29,0) DGEMM_C6X5_MKER_LOOP_PLAIN_LOC_FWD(18,19,20,24,25,26,0) DGEMM_C6X5_MKER_LOOP_PLAIN_LOC_FWD(21,22,23,27,28,29,0) BRANCH(K_MKER_LOOP) // // Final microkernel loop. LABEL(FIN_MKER_LOOP) DGEMM_C6X5_MKER_LOOP_PLAIN_LOC(15,16,17,27,28,29,0,xzr,-1,xzr,-1,noload) " ldr q27, [x1, #16*0] \n\t" " ldr q28, [x1, #16*1] \n\t" " ldr d29, [x1, #16*2] \n\t" " add x1, x1, x3 \n\t" DGEMM_C6X5_MKER_LOOP_PLAIN_LOC(18,19,20,24,25,26,0,xzr,-1,xzr,-1,noload) DGEMM_C6X5_MKER_LOOP_PLAIN_LOC(21,22,23,27,28,29,0,xzr,-1,xzr,-1,noload) // // Loops left behind microkernels. LABEL(K_LEFT_LOOP) " cmp x8, #0 \n\t" // End of exec. BEQ(WRITE_MEM_PREP) " mov x14, x0 \n\t" // Load A col. " ld1 {v15.d}[0], [x14], x9 \n\t" " ld1 {v15.d}[1], [x14], x9 \n\t" " ld1 {v16.d}[0], [x14], x9 \n\t" " ld1 {v16.d}[1], [x14], x9 \n\t" " ld1 {v17.d}[0], [x14], x9 \n\t" " ld1 {v17.d}[1], [x14], x9 \n\t" " add x0, x0, x2 \n\t" " ldr q24, [x1, #16*0] \n\t" // Load B row. " ldr q25, [x1, #16*1] \n\t" " ldr d26, [x1, #16*2] \n\t" " add x1, x1, x3 \n\t" " sub x8, x8, #1 \n\t" DGEMM_C6X5_MKER_LOOP_PLAIN_LOC(15,16,17,24,25,26,0,xzr,-1,xzr,-1,noload) BRANCH(K_LEFT_LOOP) // // Scale and write to memory. LABEL(WRITE_MEM_PREP) " ldr x4, %[alpha] \n\t" // Load alpha & beta (address). " ldr x8, %[beta] \n\t" " ld1r {v30.2d}, [x4] \n\t" // Load alpha & beta. " ld1r {v31.2d}, [x8] \n\t" " fmov d26, #1.0 \n\t" " fcmp d30, d26 \n\t" BEQ(UNIT_ALPHA) DSCALE5V(0,1,2,3,4,30,0) DSCALE5V(5,6,7,8,9,30,0) DSCALE5V(10,11,12,13,14,30,0) LABEL(UNIT_ALPHA) " \n\t" " mov x1, x5 \n\t" // C address for loading. " \n\t" // C address for storing is x5 itself. " cmp x7, #8 \n\t" // Check for column-storage. BNE(WRITE_MEM_C) // Unlike other RV kernels, here row-storage of C requires // in-register transpose. " trn1 v15.2d, v0.2d, v1.2d \n\t" " trn2 v16.2d, v0.2d, v1.2d \n\t" " trn1 v17.2d, v2.2d, v3.2d \n\t" " trn2 v18.2d, v2.2d, v3.2d \n\t" " \n\t" " trn1 v19.2d, v5.2d, v6.2d \n\t" " trn2 v20.2d, v5.2d, v6.2d \n\t" " trn1 v21.2d, v7.2d, v8.2d \n\t" " trn2 v22.2d, v7.2d, v8.2d \n\t" " \n\t" " trn1 v23.2d, v10.2d, v11.2d \n\t" " trn2 v24.2d, v10.2d, v11.2d \n\t" " trn1 v25.2d, v12.2d, v13.2d \n\t" " trn2 v26.2d, v12.2d, v13.2d \n\t" " \n\t" " fcmp d31, #0.0 \n\t" BEQ(ZERO_BETA_R) DLOADC_2PHV_R_FWD(0,1,28,0,x1,0,x6,x8) DLOADC_2PHV_R_FWD(2,3,28,1,x1,0,x6,x8) DLOADC_2PHV_R_FWD(5,6,29,0,x1,0,x6,x8) DLOADC_2PHV_R_FWD(7,8,29,1,x1,0,x6,x8) DLOADC_2PHV_R_FWD(10,11,30,0,x1,0,x6,x8) DLOADC_2PHV_R_FWD(12,13,30,1,x1,0,x6,x8) DSCALEA5V(15,17,16,18,4,0,1,2,3,28,31,0) DSCALEA5V(19,21,20,22,9,5,6,7,8,29,31,0) DSCALEA5V(23,25,24,26,14,10,11,12,13,30,31,0) LABEL(ZERO_BETA_R) #ifndef __clang__ " cmp x12, #1 \n\t" BRANCH(PRFM_END_R) " prfm PLDL1KEEP, [%[a_next], #16*0] \n\t" " prfm PLDL1KEEP, [%[a_next], #16*1] \n\t" " prfm PLDL1STRM, [%[b_next], #16*0] \n\t" " prfm PLDL1STRM, [%[b_next], #16*1] \n\t" LABEL(PRFM_END_R) #endif DSTOREC_2PHV_R_FWD(15,17,4,0,x5,0,x6,x8) DSTOREC_2PHV_R_FWD(16,18,4,1,x5,0,x6,x8) DSTOREC_2PHV_R_FWD(19,21,9,0,x5,0,x6,x8) DSTOREC_2PHV_R_FWD(20,22,9,1,x5,0,x6,x8) DSTOREC_2PHV_R_FWD(23,25,14,0,x5,0,x6,x8) DSTOREC_2PHV_R_FWD(24,26,14,1,x5,0,x6,x8) BRANCH(END_WRITE_MEM) // // C storage in columns. LABEL(WRITE_MEM_C) " fcmp d31, #0.0 \n\t" BEQ(ZERO_BETA_C) DLOADC_3V_C_FWD(15,20,25,x1,0,x7) DLOADC_3V_C_FWD(16,21,26,x1,0,x7) DLOADC_3V_C_FWD(17,22,27,x1,0,x7) DLOADC_3V_C_FWD(18,23,28,x1,0,x7) DLOADC_3V_C_FWD(19,24,29,x1,0,x7) DSCALEA5V(0,1,2,3,4,15,16,17,18,19,31,0) DSCALEA5V(5,6,7,8,9,20,21,22,23,24,31,0) DSCALEA5V(10,11,12,13,14,25,26,27,28,29,31,0) LABEL(ZERO_BETA_C) #ifndef __clang__ " cmp x12, #1 \n\t" BRANCH(PRFM_END_C) " prfm PLDL1KEEP, [%[a_next], #16*0] \n\t" " prfm PLDL1KEEP, [%[a_next], #16*1] \n\t" " prfm PLDL1STRM, [%[b_next], #16*0] \n\t" " prfm PLDL1STRM, [%[b_next], #16*1] \n\t" LABEL(PRFM_END_C) #endif DSTOREC_3V_C_FWD(0,5,10,x5,0,x7) DSTOREC_3V_C_FWD(1,6,11,x5,0,x7) DSTOREC_3V_C_FWD(2,7,12,x5,0,x7) DSTOREC_3V_C_FWD(3,8,13,x5,0,x7) DSTOREC_3V_C_FWD(4,9,14,x5,0,x7) // // End of this microkernel. LABEL(END_WRITE_MEM) " \n\t" " subs x12, x12, #1 \n\t" BEQ(END_EXEC) " \n\t" " mov x8, #6 \n\t" " madd x13, x6, x8, x13 \n\t" // Forward C's base address to the next logic panel. " add x10, x10, x11 \n\t" // Forward A's base address to the next logic panel. BRANCH(MILLIKER_MLOOP) // // End of execution. LABEL(END_EXEC) : : [a] "m" (a), [b] "m" (b), [c] "m" (c), [rs_a] "m" (rs_a), [cs_a] "m" (cs_a), [ps_a] "m" (ps_a), [rs_b] "m" (rs_b), [rs_c] "m" (rs_c), [cs_c] "m" (cs_c), // In Clang, even "m"-passed parameter takes 1 register. // Have to disable prefetching to pass compilation. #ifndef __clang__ [a_next] "r" (a_next), [b_next] "r" (b_next), #endif [m_iter] "m" (m_iter), [k_mker] "m" (k_mker), [k_left] "m" (k_left), [alpha] "m" (alpha), [beta] "m" (beta) : "x0", "x1", "x2", "x3", "x4", "x5", "x6", "x7", "x8", "x9", "x10","x11","x12","x13","x14", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10","v11","v12","v13","v14","v15", "v16","v17","v18","v19","v20","v21","v22","v23", "v24","v25","v26","v27","v28","v29","v30","v31" ); consider_edge_cases: // Forward address. a = ( double* )a + m_iter * ps_a; c = ( double* )c + m_iter * 6 * rs_c; auxinfo_t data_d6x4mn = *data; bli_auxinfo_set_ps_b( 4 * cs_b0, &data_d6x4mn ); bli_dgemmsup_rv_armv8a_int_6x4mn ( conja, conjb, m_left, 5, k0, alpha, a, rs_a0, cs_a0, b, rs_b0, cs_b0, beta, c, rs_c0, cs_c0, &data_d6x4mn, cntx ); }