// Copyright 2019 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include "xnnpack/assembly.h" # void xnn_f16_igemm_minmax_ukernel_6x16__asm_aarch64_neonfp16arith_cortex_a55r0( # size_t mr, x0 # size_t nc, x1 # size_t kc, x2 / x0 # size_t ks, x3 / x9 # const void** restrict a, x4 # const void* restrict w, x5 # uint8_t* restrict c, x6 # size_t cm_stride, x7 # size_t cn_stride, [sp] -> (x0) # size_t a_offset, [sp + 8] -> x11 # const void* zero, [sp + 16] -> x12 # const xnn_f16_minmax_params params [sp + 24] -> (x8) # d8-d15, x19-x30 need to be preserved if used. x18 is reserved by the OS. // Register usage // A0 x14 v0 v3 // A1 x15 v0[1] v3[1] // A2 x20 v1 v4 // A3 x21 v1[1] v4[1] // A4 x22 v2 v5 // A5 x23 v2[1] v5[1] // B x5 v12 v13 v14 v15 second set of B // B v16 v17 v18 v19 first set // C0 x6 v20 v21 // C1 x16 v22 v23 // C2 x17 v24 v25 // C3 x10 v26 v27 // C4 x13 v28 v29 // C5 x7 v30 v31 // clamp v6, (v4), (v5) // unused v7 v8 v9 v10 v11 // temporary vector shadow register x8 BEGIN_FUNCTION xnn_f16_igemm_minmax_ukernel_6x16__asm_aarch64_neonfp16arith_cortex_a55r0 # Load zero, params pointer LDP x12, x8, [sp, 16] # Clamp C pointers CMP x0, 2 // if mr < 2 ADD x16, x6, x7 // c1 = c0 + cm_stride CSEL x16, x6, x16, LO // c1 = c0 ADD x17, x16, x7 // c2 = c1 + cm_stride // if mr <= 2 CSEL x17, x16, x17, LS // c2 = c1 # Load params LDR s6, [x8] CMP x0, 4 // if mr < 4 ADD x10, x17, x7 // c3 = c2 + cm_stride CSEL x10, x17, x10, LO // c3 = c2 ADD x13, x10, x7 // c4 = c3 + cm_stride // if mr <= 4 CSEL x13, x10, x13, LS // c4 = c3 CMP x0, 6 // if mr < 6 ADD x7, x13, x7 // c5 = c4 + cm_stride CSEL x7, x13, x7, LO // c5 = c4 # Load a_offset LDR x11, [sp, 8] # Save x20-x23, d12-d15 on stack STP d12, d13, [sp, -64]! STP d14, d15, [sp, 16] STP x20, x21, [sp, 32] STP x22, x23, [sp, 48] 0: # Load initial bias from w into accumulators LDP q20, q21, [x5], 32 MOV x9, x3 // p = ks MOV v22.16b, v20.16b MOV v23.16b, v21.16b MOV v24.16b, v20.16b MOV v25.16b, v21.16b MOV v26.16b, v20.16b MOV v27.16b, v21.16b MOV v28.16b, v20.16b MOV v29.16b, v21.16b MOV v30.16b, v20.16b MOV v31.16b, v21.16b 1: # Load next 6 A pointers LDP x14, x15, [x4], 16 LDP x20, x21, [x4], 16 LDP x22, x23, [x4], 16 CMP x14, x12 // if a0 == zero ADD x14, x14, x11 // a0 += a_offset CSEL x14, x12, x14, EQ // a0 = zero, else += a0 + a_offset CMP x15, x12 // if a1 == zero ADD x15, x15, x11 // a1 += a_offset CSEL x15, x12, x15, EQ // a1 = zero, else += a1 + a_offset CMP x20, x12 // if a2 == zero ADD x20, x20, x11 // a2 += a_offset CSEL x20, x12, x20, EQ // a2 = zero, else += a2 + a_offset CMP x21, x12 // if a3 == zero ADD x21, x21, x11 // a3 += a_offset CSEL x21, x12, x21, EQ // a3 = zero, else += a3 + a_offset CMP x22, x12 // if a4 == zero ADD x22, x22, x11 // a4 += a_offset CSEL x22, x12, x22, EQ // a4 = zero, else += a4 + a_offset CMP x23, x12 // if a5 == zero ADD x23, x23, x11 // a5 += a_offset CSEL x23, x12, x23, EQ // a5 = zero, else += a5 + a_offset # Is there at least 4 halffloats (8 bytes) for prologue + epilogue? SUBS x0, x2, 8 // k = kc - 8 B.LO 5f # Prologue - First group loads, no FMA LDR s0, [x14], 4 // A0 LDP q16, q17, [x5], 32 // B LDR s1, [x20], 4 // A2 LDR s2, [x22], 4 // A4 LD1 {v0.s}[2], [x15], 4 // A1 LD1 {v1.s}[2], [x21], 4 // A3 LD1 {v2.s}[2], [x23], 4 // A5 LDR q18, [x5], 16 LDR d19, [x5], 8 LDR x8, [x5], 8 // ins is in BLOCK 0 SUBS x0, x0, 8 # Is there at least 4 halffloats (8 bytes) for main loop? B.LO 3f .p2align 3 # Main loop - 4 halffloats of A (8 bytes) # 48 FMA + 12 LD32 A + 8 LDR B 2: # First group of 24 FMA, Second group loads # BLOCK 0 LDR s3, [x14], 4 // A0 INS v19.d[1], x8 // B from second group FMLA v20.8h, v16.8h, v0.h[0] LDR w8, [x15], 4 // A1 FMLA v22.8h, v16.8h, v0.h[4] FMLA v24.8h, v16.8h, v1.h[0] # BLOCK 1 LDR d12, [x5] INS v3.d[1], x8 // A1 ins FMLA v26.8h, v16.8h, v1.h[4] LDR x8, [x5, 8] // B FMLA v28.8h, v16.8h, v2.h[0] FMLA v30.8h, v16.8h, v2.h[4] # BLOCK 2 LDR s4, [x20], 4 // A2 INS v12.d[1], x8 // B ins FMLA v21.8h, v17.8h, v0.h[0] LDR w8, [x21], 4 // A3 FMLA v23.8h, v17.8h, v0.h[4] FMLA v25.8h, v17.8h, v1.h[0] # BLOCK 3 LDR s5, [x22], 4 // A4 INS v4.d[1], x8 // A3 ins FMLA v27.8h, v17.8h, v1.h[4] LDR w8, [x23], 4 // A5 FMLA v29.8h, v17.8h, v2.h[0] FMLA v31.8h, v17.8h, v2.h[4] # BLOCK 4 LDR d13, [x5, 16] INS v5.d[1], x8 // A5 ins FMLA v20.8h, v18.8h, v0.h[1] LDR x8, [x5, 24] FMLA v22.8h, v18.8h, v0.h[5] FMLA v24.8h, v18.8h, v1.h[1] # BLOCK 5 LDR d14, [x5, 32] INS v13.d[1], x8 // B FMLA v26.8h, v18.8h, v1.h[5] LDR x8, [x5, 40] FMLA v28.8h, v18.8h, v2.h[1] FMLA v30.8h, v18.8h, v2.h[5] # BLOCK 6 LDR d15, [x5, 48] INS v14.d[1], x8 // B FMLA v21.8h, v19.8h, v0.h[1] LDR x8, [x5, 56] FMLA v23.8h, v19.8h, v0.h[5] FMLA v25.8h, v19.8h, v1.h[1] # BLOCK 7 INS v15.d[1], x8 FMLA v27.8h, v19.8h, v1.h[5] FMLA v29.8h, v19.8h, v2.h[1] FMLA v31.8h, v19.8h, v2.h[5] # Second group of 24 FMA, First group of loads # BLOCK 0 LDR s0, [x14], 4 // A0 FMLA v20.8h, v12.8h, v3.h[0] LDR w8, [x15], 4 // A1 FMLA v22.8h, v12.8h, v3.h[4] FMLA v24.8h, v12.8h, v4.h[0] # BLOCK 1 LDR d16, [x5, 64] INS v0.d[1], x8 // A1 ins FMLA v26.8h, v12.8h, v4.h[4] LDR x8, [x5, 72] // B FMLA v28.8h, v12.8h, v5.h[0] FMLA v30.8h, v12.8h, v5.h[4] # BLOCK 2 LDR s1, [x20], 4 // A2 INS v16.d[1], x8 // B FMLA v21.8h, v13.8h, v3.h[0] LDR w8, [x21], 4 // A3 FMLA v23.8h, v13.8h, v3.h[4] FMLA v25.8h, v13.8h, v4.h[0] # BLOCK 3 LDR s2, [x22], 4 // A4 INS v1.d[1], x8 // A3 ins FMLA v27.8h, v13.8h, v4.h[4] LDR w8, [x23], 4 // A5 FMLA v29.8h, v13.8h, v5.h[0] FMLA v31.8h, v13.8h, v5.h[4] # BLOCK 4 LDR d17, [x5, 80] INS v2.d[1], x8 // A5 ins FMLA v20.8h, v14.8h, v3.h[1] LDR x8, [x5, 88] FMLA v22.8h, v14.8h, v3.h[5] FMLA v24.8h, v14.8h, v4.h[1] # BLOCK 5 LDR d18, [x5, 96] INS v17.d[1], x8 // B FMLA v26.8h, v14.8h, v4.h[5] LDR x8, [x5, 104] FMLA v28.8h, v14.8h, v5.h[1] FMLA v30.8h, v14.8h, v5.h[5] # BLOCK 6 LDR d19, [x5, 112] INS v18.d[1], x8 // B FMLA v21.8h, v15.8h, v3.h[1] LDR x8, [x5, 120] FMLA v23.8h, v15.8h, v3.h[5] FMLA v25.8h, v15.8h, v4.h[1] # BLOCK 7 SUBS x0, x0, 8 // LDR lands here FMLA v27.8h, v15.8h, v4.h[5] FMLA v29.8h, v15.8h, v5.h[1] ADD x5, x5, 128 FMLA v31.8h, v15.8h, v5.h[5] B.HS 2b # Epilogue - 4 halffloats of A (8 bytes) # 48 FMA + 12 LD32 A + 8 LDR B 3: # First group of 24 FMA, Second group loads # BLOCK 0 LDR s3, [x14], 4 // A0 INS v19.d[1], x8 // B from second group FMLA v20.8h, v16.8h, v0.h[0] LDR w8, [x15], 4 // A1 FMLA v22.8h, v16.8h, v0.h[4] FMLA v24.8h, v16.8h, v1.h[0] # BLOCK 1 LDR d12, [x5] INS v3.d[1], x8 // A1 ins FMLA v26.8h, v16.8h, v1.h[4] LDR x8, [x5, 8] // B FMLA v28.8h, v16.8h, v2.h[0] FMLA v30.8h, v16.8h, v2.h[4] # BLOCK 2 LDR s4, [x20], 4 // A2 INS v12.d[1], x8 // B ins FMLA v21.8h, v17.8h, v0.h[0] LDR w8, [x21], 4 // A3 FMLA v23.8h, v17.8h, v0.h[4] FMLA v25.8h, v17.8h, v1.h[0] # BLOCK 3 LDR s5, [x22], 4 // A4 INS v4.d[1], x8 // A3 ins FMLA v27.8h, v17.8h, v1.h[4] LDR w8, [x23], 4 // A5 FMLA v29.8h, v17.8h, v2.h[0] FMLA v31.8h, v17.8h, v2.h[4] # BLOCK 4 LDR d13, [x5, 16] INS v5.d[1], x8 // A5 ins FMLA v20.8h, v18.8h, v0.h[1] LDR x8, [x5, 24] FMLA v22.8h, v18.8h, v0.h[5] FMLA v24.8h, v18.8h, v1.h[1] # BLOCK 5 LDR d14, [x5, 32] INS v13.d[1], x8 // B FMLA v26.8h, v18.8h, v1.h[5] LDR x8, [x5, 40] FMLA v28.8h, v18.8h, v2.h[1] FMLA v30.8h, v18.8h, v2.h[5] # BLOCK 6 LDR d15, [x5, 48] INS v14.d[1], x8 // B FMLA v21.8h, v19.8h, v0.h[1] LDR x8, [x5, 56] FMLA v23.8h, v19.8h, v0.h[5] FMLA v25.8h, v19.8h, v1.h[1] # BLOCK 7 INS v15.d[1], x8 // B FMLA v27.8h, v19.8h, v1.h[5] FMLA v29.8h, v19.8h, v2.h[1] FMLA v31.8h, v19.8h, v2.h[5] # Second group of 24 FMA, First group of loads # BLOCK 0 FMLA v20.8h, v12.8h, v3.h[0] FMLA v22.8h, v12.8h, v3.h[4] FMLA v24.8h, v12.8h, v4.h[0] # BLOCK 1 FMLA v26.8h, v12.8h, v4.h[4] FMLA v28.8h, v12.8h, v5.h[0] FMLA v30.8h, v12.8h, v5.h[4] # BLOCK 2 FMLA v21.8h, v13.8h, v3.h[0] FMLA v23.8h, v13.8h, v3.h[4] FMLA v25.8h, v13.8h, v4.h[0] # BLOCK 3 FMLA v27.8h, v13.8h, v4.h[4] FMLA v29.8h, v13.8h, v5.h[0] FMLA v31.8h, v13.8h, v5.h[4] # BLOCK 4 FMLA v20.8h, v14.8h, v3.h[1] FMLA v22.8h, v14.8h, v3.h[5] FMLA v24.8h, v14.8h, v4.h[1] # BLOCK 5 FMLA v26.8h, v14.8h, v4.h[5] FMLA v28.8h, v14.8h, v5.h[1] FMLA v30.8h, v14.8h, v5.h[5] TST x0, 7 # BLOCK 6 FMLA v21.8h, v15.8h, v3.h[1] FMLA v23.8h, v15.8h, v3.h[5] FMLA v25.8h, v15.8h, v4.h[1] ADD x5, x5, 64 # BLOCK 7 FMLA v27.8h, v15.8h, v4.h[5] FMLA v29.8h, v15.8h, v5.h[1] FMLA v31.8h, v15.8h, v5.h[5] # Is there a remainder?- 2 halffloats of A (4 bytes) or less B.NE 5f 4: # ks loop SUBS x9, x9, 48 // ks -= MR * sizeof(void*) B.HI 1b # Clamp DUP v4.8h, v6.h[0] DUP v5.8h, v6.h[1] LDR x0, [sp, 64] // cn_stride FMAX v20.8h, v20.8h, v4.8h FMAX v21.8h, v21.8h, v4.8h FMAX v22.8h, v22.8h, v4.8h FMAX v23.8h, v23.8h, v4.8h FMAX v24.8h, v24.8h, v4.8h FMAX v25.8h, v25.8h, v4.8h FMAX v26.8h, v26.8h, v4.8h FMAX v27.8h, v27.8h, v4.8h FMAX v28.8h, v28.8h, v4.8h FMAX v29.8h, v29.8h, v4.8h FMAX v30.8h, v30.8h, v4.8h FMAX v31.8h, v31.8h, v4.8h SUBS x1, x1, 16 FMIN v20.8h, v20.8h, v5.8h FMIN v21.8h, v21.8h, v5.8h FMIN v22.8h, v22.8h, v5.8h FMIN v23.8h, v23.8h, v5.8h FMIN v24.8h, v24.8h, v5.8h FMIN v25.8h, v25.8h, v5.8h FMIN v26.8h, v26.8h, v5.8h FMIN v27.8h, v27.8h, v5.8h FMIN v28.8h, v28.8h, v5.8h FMIN v29.8h, v29.8h, v5.8h FMIN v30.8h, v30.8h, v5.8h FMIN v31.8h, v31.8h, v5.8h # Store full 6 x 16 B.LO 7f ST1 {v30.16b, v31.16b}, [x7], x0 ST1 {v28.16b, v29.16b}, [x13], x0 ST1 {v26.16b, v27.16b}, [x10], x0 ST1 {v24.16b, v25.16b}, [x17], x0 ST1 {v22.16b, v23.16b}, [x16], x0 ST1 {v20.16b, v21.16b}, [x6], x0 SUB x4, x4, x3 // a -= ks # nc loop B.HI 0b # Restore x20-x23, d12-d15 from stack LDP x22, x23, [sp, 48] LDP x20, x21, [sp, 32] LDP d14, d15, [sp, 16] LDP d12, d13, [sp], 64 RET 5: # Is there a remainder?- 2 halffloats of A (4 bytes) TBZ x0, 2, 6f # Remainder- 2 halffloats of A (4 bytes) LDR s0, [x14], 4 // A0 LDP q16, q17, [x5], 32 // B LDR s1, [x20], 4 // A2 LDR s2, [x22], 4 // A4 LD1 {v0.s}[2], [x15], 4 // A1 LD1 {v1.s}[2], [x21], 4 // A3 LD1 {v2.s}[2], [x23], 4 // A5 LDR q18, [x5], 16 LDR q19, [x5], 16 FMLA v20.8h, v16.8h, v0.h[0] FMLA v22.8h, v16.8h, v0.h[4] FMLA v24.8h, v16.8h, v1.h[0] FMLA v26.8h, v16.8h, v1.h[4] FMLA v28.8h, v16.8h, v2.h[0] FMLA v30.8h, v16.8h, v2.h[4] FMLA v21.8h, v17.8h, v0.h[0] FMLA v23.8h, v17.8h, v0.h[4] FMLA v25.8h, v17.8h, v1.h[0] FMLA v27.8h, v17.8h, v1.h[4] FMLA v29.8h, v17.8h, v2.h[0] FMLA v31.8h, v17.8h, v2.h[4] FMLA v20.8h, v18.8h, v0.h[1] FMLA v22.8h, v18.8h, v0.h[5] FMLA v24.8h, v18.8h, v1.h[1] FMLA v26.8h, v18.8h, v1.h[5] FMLA v28.8h, v18.8h, v2.h[1] FMLA v30.8h, v18.8h, v2.h[5] FMLA v21.8h, v19.8h, v0.h[1] FMLA v23.8h, v19.8h, v0.h[5] FMLA v25.8h, v19.8h, v1.h[1] FMLA v27.8h, v19.8h, v1.h[5] FMLA v29.8h, v19.8h, v2.h[1] FMLA v31.8h, v19.8h, v2.h[5] # Is there a remainder?- 1 halffloat of A (2 bytes) TBZ x0, 1, 4b 6: # Remainder- 1 halffloat of A (2 bytes) LDR h0, [x14], 2 // A0 LDP q16, q17, [x5], 32 // B LDR h1, [x20], 2 // A2 LDR h2, [x22], 2 // A4 LD1 {v0.h}[4], [x15], 2 // A1 LD1 {v1.h}[4], [x21], 2 // A3 LD1 {v2.h}[4], [x23], 2 // A5 FMLA v20.8h, v16.8h, v0.h[0] FMLA v22.8h, v16.8h, v0.h[4] FMLA v24.8h, v16.8h, v1.h[0] FMLA v26.8h, v16.8h, v1.h[4] FMLA v28.8h, v16.8h, v2.h[0] FMLA v30.8h, v16.8h, v2.h[4] FMLA v21.8h, v17.8h, v0.h[0] FMLA v23.8h, v17.8h, v0.h[4] FMLA v25.8h, v17.8h, v1.h[0] FMLA v27.8h, v17.8h, v1.h[4] FMLA v29.8h, v17.8h, v2.h[0] FMLA v31.8h, v17.8h, v2.h[4] B 4b # Store odd width 7: TBZ x1, 3, 8f STR q30, [x7], 16 MOV v30.16b, v31.16b STR q28, [x13], 16 MOV v28.16b, v29.16b STR q26, [x10], 16 MOV v26.16b, v27.16b STR q24, [x17], 16 MOV v24.16b, v25.16b STR q22, [x16], 16 MOV v22.16b, v23.16b STR q20, [x6], 16 MOV v20.16b, v21.16b 8: TBZ x1, 2, 9f STR d30, [x7], 8 STR d28, [x13], 8 DUP d30, v30.d[1] DUP d28, v28.d[1] STR d26, [x10], 8 STR d24, [x17], 8 DUP d26, v26.d[1] DUP d24, v24.d[1] STR d22, [x16], 8 STR d20, [x6], 8 DUP d22, v22.d[1] DUP d20, v20.d[1] 9: TBZ x1, 1, 10f STR s30, [x7], 4 STR s28, [x13], 4 DUP s30, v30.s[1] DUP s28, v28.s[1] STR s26, [x10], 4 STR s24, [x17], 4 DUP s26, v26.s[1] DUP s24, v24.s[1] STR s22, [x16], 4 STR s20, [x6], 4 DUP s22, v22.s[1] DUP s20, v20.s[1] 10: TBZ x1, 0, 11f STR h30, [x7] STR h28, [x13] STR h26, [x10] STR h24, [x17] STR h22, [x16] STR h20, [x6] 11: # Restore x20-x23, d12-d15 from stack LDP x22, x23, [sp, 48] LDP x20, x21, [sp, 32] LDP d14, d15, [sp, 16] LDP d12, d13, [sp], 64 RET END_FUNCTION xnn_f16_igemm_minmax_ukernel_6x16__asm_aarch64_neonfp16arith_cortex_a55r0 #ifdef __ELF__ .section ".note.GNU-stack","",%progbits #endif