/* * Copyright (c) 2021-2023 Arm Limited. * * SPDX-License-Identifier: MIT * * Permission is hereby granted, free of charge, to any person obtaining a copy * of this software and associated documentation files (the "Software"), to * deal in the Software without restriction, including without limitation the * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or * sell copies of the Software, and to permit persons to whom the Software is * furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice shall be included in all * copies or substantial portions of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ #include "pooling.hpp" #include #include #if defined(ARM_COMPUTE_ENABLE_SVE) namespace arm_conv { namespace pooling { void sve_s8q_nhwc_max_generic_depthfirst_impl( const uint64_t, const uint64_t n_valid_cells, uint64_t n_channels, const int8_t *const *const inptrs, int8_t *outptr, const Requantize32 &qp ) { __asm__ __volatile__( "mov x9, #0x0\n" "cntb x28\n" "cntb x27, ALL, MUL #2\n" "cntb x26, ALL, MUL #3\n" "whilelt p4.b, x9, %x[n_channels]\n" "whilelt p3.b, x28, %x[n_channels]\n" "whilelt p2.b, x27, %x[n_channels]\n" "whilelt p1.b, x26, %x[n_channels]\n" "ptrue p0.b\n" "b.none 7f\n" "1:" // 4-vectors of channels "lsr x25, %x[n_valid_cells], #0x2\n" "mov z8.b, #0x80\n" "mov z7.b, #0x80\n" "mov x24, %x[inptrs]\n" "mov z6.b, #0x80\n" "mov z5.b, #0x80\n" "cbz x25, 4f\n" "ldp x23, x22, [x24, #0x0]\n" "ldp x21, x20, [x24, #0x10]\n" "subs x25, x25, #0x1\n" "add x24, x24, #0x20\n" "ld1b { z4.b }, p4/Z, [x23, x9]\n" "ld1b { z3.b }, p4/Z, [x22, x9]\n" "ld1b { z2.b }, p4/Z, [x21, x9]\n" "ld1b { z1.b }, p4/Z, [x20, x9]\n" "ld1b { z0.b }, p3/Z, [x23, x28]\n" "ld1b { z31.b }, p3/Z, [x22, x28]\n" "ld1b { z22.b }, p3/Z, [x21, x28]\n" "ld1b { z30.b }, p3/Z, [x20, x28]\n" "ld1b { z29.b }, p2/Z, [x23, x27]\n" "ld1b { z28.b }, p2/Z, [x22, x27]\n" "ld1b { z21.b }, p2/Z, [x21, x27]\n" "ld1b { z27.b }, p2/Z, [x20, x27]\n" "ld1b { z26.b }, p1/Z, [x23, x26]\n" "ld1b { z25.b }, p1/Z, [x22, x26]\n" "ld1b { z20.b }, p1/Z, [x21, x26]\n" "ld1b { z24.b }, p1/Z, [x20, x26]\n" "beq 3f\n" "2:" // 4-vectors of channels: 4 inputs loop "movprfx z19, z4\n smax z19.b, p0/M, z19.b, z3.b\n" "movprfx z23, z2\n smax z23.b, p0/M, z23.b, z1.b\n" "ldp x23, x22, [x24, #0x0]\n" "ldp x21, x20, [x24, #0x10]\n" "movprfx z18, z0\n smax z18.b, p0/M, z18.b, z31.b\n" "smax z22.b, p0/M, z22.b, z30.b\n" "ld1b { z4.b }, p4/Z, [x23, x9]\n" "ld1b { z3.b }, p4/Z, [x22, x9]\n" "movprfx z17, z29\n smax z17.b, p0/M, z17.b, z28.b\n" "smax z21.b, p0/M, z21.b, z27.b\n" "ld1b { z2.b }, p4/Z, [x21, x9]\n" "ld1b { z1.b }, p4/Z, [x20, x9]\n" "movprfx z16, z26\n smax z16.b, p0/M, z16.b, z25.b\n" "smax z20.b, p0/M, z20.b, z24.b\n" "ld1b { z0.b }, p3/Z, [x23, x28]\n" "ld1b { z31.b }, p3/Z, [x22, x28]\n" "smax z19.b, p0/M, z19.b, z23.b\n" "smax z18.b, p0/M, z18.b, z22.b\n" "ld1b { z22.b }, p3/Z, [x21, x28]\n" "ld1b { z30.b }, p3/Z, [x20, x28]\n" "smax z17.b, p0/M, z17.b, z21.b\n" "smax z16.b, p0/M, z16.b, z20.b\n" "ld1b { z29.b }, p2/Z, [x23, x27]\n" "ld1b { z28.b }, p2/Z, [x22, x27]\n" "subs x25, x25, #0x1\n" "smax z8.b, p0/M, z8.b, z19.b\n" "ld1b { z21.b }, p2/Z, [x21, x27]\n" "ld1b { z27.b }, p2/Z, [x20, x27]\n" "smax z7.b, p0/M, z7.b, z18.b\n" "smax z6.b, p0/M, z6.b, z17.b\n" "ld1b { z26.b }, p1/Z, [x23, x26]\n" "ld1b { z25.b }, p1/Z, [x22, x26]\n" "smax z5.b, p0/M, z5.b, z16.b\n" "add x24, x24, #0x20\n" "ld1b { z20.b }, p1/Z, [x21, x26]\n" "ld1b { z24.b }, p1/Z, [x20, x26]\n" "bgt 2b\n" "3:" // 4-vectors of channels: 4 inputs tail "movprfx z19, z4\n smax z19.b, p0/M, z19.b, z3.b\n" "movprfx z23, z2\n smax z23.b, p0/M, z23.b, z1.b\n" "movprfx z18, z0\n smax z18.b, p0/M, z18.b, z31.b\n" "smax z22.b, p0/M, z22.b, z30.b\n" "movprfx z17, z29\n smax z17.b, p0/M, z17.b, z28.b\n" "smax z21.b, p0/M, z21.b, z27.b\n" "movprfx z16, z26\n smax z16.b, p0/M, z16.b, z25.b\n" "smax z20.b, p0/M, z20.b, z24.b\n" "smax z19.b, p0/M, z19.b, z23.b\n" "smax z18.b, p0/M, z18.b, z22.b\n" "smax z17.b, p0/M, z17.b, z21.b\n" "smax z16.b, p0/M, z16.b, z20.b\n" "smax z8.b, p0/M, z8.b, z19.b\n" "smax z7.b, p0/M, z7.b, z18.b\n" "smax z6.b, p0/M, z6.b, z17.b\n" "smax z5.b, p0/M, z5.b, z16.b\n" "4:" // 4-vectors of channels: After loop "ands x21, %x[n_valid_cells], #0x3\n" "beq 6f\n" "5:" // 4-vectors of channels: Single input loop "ldr x20, [x24], #0x8\n" "ld1b { z16.b }, p4/Z, [x20, x9]\n" "subs x21, x21, #0x1\n" "smax z8.b, p0/M, z8.b, z16.b\n" "ld1b { z17.b }, p3/Z, [x20, x28]\n" "ld1b { z16.b }, p2/Z, [x20, x27]\n" "smax z7.b, p0/M, z7.b, z17.b\n" "smax z6.b, p0/M, z6.b, z16.b\n" "ld1b { z16.b }, p1/Z, [x20, x26]\n" "smax z5.b, p0/M, z5.b, z16.b\n" "bgt 5b\n" "6:" // 4-vectors of channels: Single input loop: End ".inst 0x4508a111 // sshllb z17.h, z8.b, #0x0\n" ".inst 0x4508a517 // sshllt z23.h, z8.b, #0x0\n" "add x20, %x[quant_params], %[offsetof_qp_per_layer_left_shift]\n" "ld1rw { z4.s }, p0/Z, [x20]\n" ".inst 0x4508a0f6 // sshllb z22.h, z7.b, #0x0\n" ".inst 0x4508a4f5 // sshllt z21.h, z7.b, #0x0\n" "add x20, %x[quant_params], %[offsetof_qp_per_layer_mul]\n" "ld1rw { z3.s }, p0/Z, [x20]\n" ".inst 0x4508a0d4 // sshllb z20.h, z6.b, #0x0\n" ".inst 0x4508a4d3 // sshllt z19.h, z6.b, #0x0\n" "add x20, %x[quant_params], %[offsetof_qp_per_layer_right_shift]\n" "ld1rw { z2.s }, p0/Z, [x20]\n" ".inst 0x4508a0b2 // sshllb z18.h, z5.b, #0x0\n" ".inst 0x4508a4b0 // sshllt z16.h, z5.b, #0x0\n" ".inst 0x4510a221 // sshllb z1.s, z17.h, #0x0\n" ".inst 0x4510a631 // sshllt z17.s, z17.h, #0x0\n" ".inst 0x44828081 // srshl z1.s, p0/M, z1.s, z4.s\n" ".inst 0x44828091 // srshl z17.s, p0/M, z17.s, z4.s\n" ".inst 0x4510a2e0 // sshllb z0.s, z23.h, #0x0\n" ".inst 0x4510a6ff // sshllt z31.s, z23.h, #0x0\n" ".inst 0x44828080 // srshl z0.s, p0/M, z0.s, z4.s\n" ".inst 0x4482809f // srshl z31.s, p0/M, z31.s, z4.s\n" ".inst 0x4510a2de // sshllb z30.s, z22.h, #0x0\n" ".inst 0x4510a6dd // sshllt z29.s, z22.h, #0x0\n" ".inst 0x4482809e // srshl z30.s, p0/M, z30.s, z4.s\n" ".inst 0x4482809d // srshl z29.s, p0/M, z29.s, z4.s\n" ".inst 0x4510a2bc // sshllb z28.s, z21.h, #0x0\n" ".inst 0x4510a6bb // sshllt z27.s, z21.h, #0x0\n" ".inst 0x4482809c // srshl z28.s, p0/M, z28.s, z4.s\n" ".inst 0x4482809b // srshl z27.s, p0/M, z27.s, z4.s\n" ".inst 0x4510a29a // sshllb z26.s, z20.h, #0x0\n" ".inst 0x4510a699 // sshllt z25.s, z20.h, #0x0\n" ".inst 0x4482809a // srshl z26.s, p0/M, z26.s, z4.s\n" ".inst 0x44828099 // srshl z25.s, p0/M, z25.s, z4.s\n" ".inst 0x4510a278 // sshllb z24.s, z19.h, #0x0\n" ".inst 0x4510a677 // sshllt z23.s, z19.h, #0x0\n" ".inst 0x44828098 // srshl z24.s, p0/M, z24.s, z4.s\n" ".inst 0x44828097 // srshl z23.s, p0/M, z23.s, z4.s\n" ".inst 0x4510a256 // sshllb z22.s, z18.h, #0x0\n" ".inst 0x4510a655 // sshllt z21.s, z18.h, #0x0\n" ".inst 0x44828096 // srshl z22.s, p0/M, z22.s, z4.s\n" ".inst 0x44828095 // srshl z21.s, p0/M, z21.s, z4.s\n" ".inst 0x4510a214 // sshllb z20.s, z16.h, #0x0\n" ".inst 0x4510a613 // sshllt z19.s, z16.h, #0x0\n" ".inst 0x44828094 // srshl z20.s, p0/M, z20.s, z4.s\n" ".inst 0x44828093 // srshl z19.s, p0/M, z19.s, z4.s\n" ".inst 0x04a37421 // sqrdmulh z1.s, z1.s, z3.s\n" ".inst 0x04a37631 // sqrdmulh z17.s, z17.s, z3.s\n" ".inst 0x44828041 // srshl z1.s, p0/M, z1.s, z2.s\n" ".inst 0x44828051 // srshl z17.s, p0/M, z17.s, z2.s\n" ".inst 0x04a37400 // sqrdmulh z0.s, z0.s, z3.s\n" ".inst 0x04a377ff // sqrdmulh z31.s, z31.s, z3.s\n" ".inst 0x44828040 // srshl z0.s, p0/M, z0.s, z2.s\n" ".inst 0x4482805f // srshl z31.s, p0/M, z31.s, z2.s\n" ".inst 0x04a377de // sqrdmulh z30.s, z30.s, z3.s\n" ".inst 0x04a377bd // sqrdmulh z29.s, z29.s, z3.s\n" ".inst 0x4482805e // srshl z30.s, p0/M, z30.s, z2.s\n" ".inst 0x4482805d // srshl z29.s, p0/M, z29.s, z2.s\n" ".inst 0x04a3779c // sqrdmulh z28.s, z28.s, z3.s\n" ".inst 0x04a3777b // sqrdmulh z27.s, z27.s, z3.s\n" ".inst 0x4482805c // srshl z28.s, p0/M, z28.s, z2.s\n" ".inst 0x4482805b // srshl z27.s, p0/M, z27.s, z2.s\n" ".inst 0x04a3775a // sqrdmulh z26.s, z26.s, z3.s\n" ".inst 0x04a37739 // sqrdmulh z25.s, z25.s, z3.s\n" ".inst 0x4482805a // srshl z26.s, p0/M, z26.s, z2.s\n" ".inst 0x44828059 // srshl z25.s, p0/M, z25.s, z2.s\n" ".inst 0x04a37718 // sqrdmulh z24.s, z24.s, z3.s\n" ".inst 0x04a376f7 // sqrdmulh z23.s, z23.s, z3.s\n" ".inst 0x44828058 // srshl z24.s, p0/M, z24.s, z2.s\n" ".inst 0x44828057 // srshl z23.s, p0/M, z23.s, z2.s\n" ".inst 0x04a376d6 // sqrdmulh z22.s, z22.s, z3.s\n" ".inst 0x04a376b5 // sqrdmulh z21.s, z21.s, z3.s\n" ".inst 0x44828056 // srshl z22.s, p0/M, z22.s, z2.s\n" ".inst 0x44828055 // srshl z21.s, p0/M, z21.s, z2.s\n" ".inst 0x04a37694 // sqrdmulh z20.s, z20.s, z3.s\n" ".inst 0x04a37673 // sqrdmulh z19.s, z19.s, z3.s\n" ".inst 0x44828054 // srshl z20.s, p0/M, z20.s, z2.s\n" ".inst 0x44828053 // srshl z19.s, p0/M, z19.s, z2.s\n" "mov z18.s, #0x7f\n" "not z16.s, p0/M, z18.s\n" "smax z1.s, p0/M, z1.s, z16.s\n" "smax z17.s, p0/M, z17.s, z16.s\n" "smax z0.s, p0/M, z0.s, z16.s\n" "smax z31.s, p0/M, z31.s, z16.s\n" "smax z30.s, p0/M, z30.s, z16.s\n" "smax z29.s, p0/M, z29.s, z16.s\n" "smax z28.s, p0/M, z28.s, z16.s\n" "smax z27.s, p0/M, z27.s, z16.s\n" "smax z26.s, p0/M, z26.s, z16.s\n" "smax z25.s, p0/M, z25.s, z16.s\n" "smax z24.s, p0/M, z24.s, z16.s\n" "smax z23.s, p0/M, z23.s, z16.s\n" "smax z22.s, p0/M, z22.s, z16.s\n" "smax z21.s, p0/M, z21.s, z16.s\n" "smax z20.s, p0/M, z20.s, z16.s\n" "smax z19.s, p0/M, z19.s, z16.s\n" "smin z1.s, p0/M, z1.s, z18.s\n" "smin z17.s, p0/M, z17.s, z18.s\n" "smin z0.s, p0/M, z0.s, z18.s\n" "trn1 z17.h, z1.h, z17.h\n" "smin z31.s, p0/M, z31.s, z18.s\n" "smin z30.s, p0/M, z30.s, z18.s\n" "trn1 z16.h, z0.h, z31.h\n" "trn1 z16.b, z17.b, z16.b\n" "smin z29.s, p0/M, z29.s, z18.s\n" "smin z28.s, p0/M, z28.s, z18.s\n" "trn1 z17.h, z30.h, z29.h\n" "st1b { z16.b }, p4, [%x[outptr], x9]\n" "smin z27.s, p0/M, z27.s, z18.s\n" "smin z26.s, p0/M, z26.s, z18.s\n" "trn1 z16.h, z28.h, z27.h\n" "trn1 z16.b, z17.b, z16.b\n" "smin z25.s, p0/M, z25.s, z18.s\n" "smin z24.s, p0/M, z24.s, z18.s\n" "trn1 z17.h, z26.h, z25.h\n" "st1b { z16.b }, p3, [%x[outptr], x28]\n" "smin z23.s, p0/M, z23.s, z18.s\n" "smin z22.s, p0/M, z22.s, z18.s\n" "trn1 z16.h, z24.h, z23.h\n" "trn1 z16.b, z17.b, z16.b\n" "smin z21.s, p0/M, z21.s, z18.s\n" "smin z20.s, p0/M, z20.s, z18.s\n" "trn1 z17.h, z22.h, z21.h\n" "st1b { z16.b }, p2, [%x[outptr], x27]\n" "smin z19.s, p0/M, z19.s, z18.s\n" "trn1 z16.h, z20.h, z19.h\n" "trn1 z16.b, z17.b, z16.b\n" "st1b { z16.b }, p1, [%x[outptr], x26]\n" "incb x26, ALL, MUL #4\n" "whilelt p1.b, x26, %x[n_channels]\n" "incb x9, ALL, MUL #4\n" "incb x28, ALL, MUL #4\n" "incb x27, ALL, MUL #4\n" "b.any 1b\n" "7:" // Single vector of channels "whilelt p4.b, x9, %x[n_channels]\n" "b.none 14f\n" "8:" // Single vector of channels: Loop "lsr x25, %x[n_valid_cells], #0x2\n" "mov z8.b, #0x80\n" "mov x24, %x[inptrs]\n" "cbz x25, 11f\n" "ldp x23, x22, [x24, #0x0]\n" "ldp x21, x20, [x24, #0x10]\n" "subs x25, x25, #0x1\n" "add x24, x24, #0x20\n" "ld1b { z4.b }, p4/Z, [x23, x9]\n" "ld1b { z3.b }, p4/Z, [x22, x9]\n" "ld1b { z2.b }, p4/Z, [x21, x9]\n" "ld1b { z1.b }, p4/Z, [x20, x9]\n" "beq 10f\n" "9:" // Single vector of channels: Loop: 4 inputs loop "movprfx z16, z4\n smax z16.b, p0/M, z16.b, z3.b\n" "movprfx z17, z2\n smax z17.b, p0/M, z17.b, z1.b\n" "ldp x23, x22, [x24, #0x0]\n" "ldp x21, x20, [x24, #0x10]\n" "smax z16.b, p0/M, z16.b, z17.b\n" "subs x25, x25, #0x1\n" "ld1b { z4.b }, p4/Z, [x23, x9]\n" "ld1b { z3.b }, p4/Z, [x22, x9]\n" "smax z8.b, p0/M, z8.b, z16.b\n" "add x24, x24, #0x20\n" "ld1b { z2.b }, p4/Z, [x21, x9]\n" "ld1b { z1.b }, p4/Z, [x20, x9]\n" "bgt 9b\n" "10:" // Single vector of channels: Loop: 4 inputs tail "movprfx z16, z4\n smax z16.b, p0/M, z16.b, z3.b\n" "movprfx z17, z2\n smax z17.b, p0/M, z17.b, z1.b\n" "smax z16.b, p0/M, z16.b, z17.b\n" "smax z8.b, p0/M, z8.b, z16.b\n" "11:" // Single vector of channels: Loop: After loop "ands x21, %x[n_valid_cells], #0x3\n" "beq 13f\n" "12:" // Single vector of channels: Loop: Single input loop "ldr x20, [x24], #0x8\n" "ld1b { z16.b }, p4/Z, [x20, x9]\n" "subs x21, x21, #0x1\n" "smax z8.b, p0/M, z8.b, z16.b\n" "bgt 12b\n" "13:" // Single vector of channels: Loop: Single input loop: End ".inst 0x4508a111 // sshllb z17.h, z8.b, #0x0\n" ".inst 0x4508a512 // sshllt z18.h, z8.b, #0x0\n" "add x20, %x[quant_params], %[offsetof_qp_per_layer_left_shift]\n" "ld1rw { z16.s }, p0/Z, [x20]\n" ".inst 0x4510a236 // sshllb z22.s, z17.h, #0x0\n" ".inst 0x4510a635 // sshllt z21.s, z17.h, #0x0\n" "add x20, %x[quant_params], %[offsetof_qp_per_layer_mul]\n" "ld1rw { z17.s }, p0/Z, [x20]\n" ".inst 0x4510a254 // sshllb z20.s, z18.h, #0x0\n" ".inst 0x4510a653 // sshllt z19.s, z18.h, #0x0\n" ".inst 0x44828216 // srshl z22.s, p0/M, z22.s, z16.s\n" ".inst 0x44828215 // srshl z21.s, p0/M, z21.s, z16.s\n" ".inst 0x44828214 // srshl z20.s, p0/M, z20.s, z16.s\n" ".inst 0x44828213 // srshl z19.s, p0/M, z19.s, z16.s\n" ".inst 0x04b176d6 // sqrdmulh z22.s, z22.s, z17.s\n" ".inst 0x04b176b5 // sqrdmulh z21.s, z21.s, z17.s\n" "add x20, %x[quant_params], %[offsetof_qp_per_layer_right_shift]\n" "ld1rw { z16.s }, p0/Z, [x20]\n" ".inst 0x04b17694 // sqrdmulh z20.s, z20.s, z17.s\n" ".inst 0x04b17673 // sqrdmulh z19.s, z19.s, z17.s\n" "mov z18.s, #0x7f\n" ".inst 0x44828216 // srshl z22.s, p0/M, z22.s, z16.s\n" ".inst 0x44828215 // srshl z21.s, p0/M, z21.s, z16.s\n" ".inst 0x44828214 // srshl z20.s, p0/M, z20.s, z16.s\n" ".inst 0x44828213 // srshl z19.s, p0/M, z19.s, z16.s\n" "not z16.s, p0/M, z18.s\n" "smax z22.s, p0/M, z22.s, z16.s\n" "smax z21.s, p0/M, z21.s, z16.s\n" "smax z20.s, p0/M, z20.s, z16.s\n" "smax z19.s, p0/M, z19.s, z16.s\n" "smin z22.s, p0/M, z22.s, z18.s\n" "smin z21.s, p0/M, z21.s, z18.s\n" "smin z20.s, p0/M, z20.s, z18.s\n" "trn1 z17.h, z22.h, z21.h\n" "smin z19.s, p0/M, z19.s, z18.s\n" "trn1 z16.h, z20.h, z19.h\n" "trn1 z16.b, z17.b, z16.b\n" "st1b { z16.b }, p4, [%x[outptr], x9]\n" "incb x9\n" "whilelt p4.b, x9, %x[n_channels]\n" "b.any 8b\n" "14:" // End : : [inptrs] "r" (inptrs), [n_channels] "r" (n_channels), [n_valid_cells] "r" (n_valid_cells), [offsetof_qp_per_layer_left_shift] "I" (offsetof(Requantize32, per_layer_left_shift)), [offsetof_qp_per_layer_mul] "I" (offsetof(Requantize32, per_layer_mul)), [offsetof_qp_per_layer_right_shift] "I" (offsetof(Requantize32, per_layer_right_shift)), [outptr] "r" (outptr), [quant_params] "r" (&qp) : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "x9", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31" ); } } // namespace pooling } // namespace arm_conv #endif // defined(ARM_COMPUTE_ENABLE_SVE)