From d02d5edfa15ba6c04a9986a8a362a945cb38ac31 Mon Sep 17 00:00:00 2001 From: Michele Di Giorgio Date: Fri, 22 Jan 2021 09:47:04 +0000 Subject: Integrate improved CPU depthwise convolution kernels * Replace assembly kernels for depthwise convolution with more optimized ones. * Add int8 assembly kernels. * Fix implicit padding on optimized kernels Resolves: COMPMID-3867, COMPMID-4361 Change-Id: I0b0867e05f61be4f368f62190d55e14d0ab3ebf2 Signed-off-by: Michele Di Giorgio Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5622 Tested-by: Arm Jenkins Reviewed-by: Georgios Pinitas --- .../generic_direct.cpp | 478 +++++++++++++++++++++ 1 file changed, 478 insertions(+) create mode 100644 src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp (limited to 'src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp') diff --git a/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp b/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp new file mode 100644 index 0000000000..aebf0bf7ac --- /dev/null +++ b/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp @@ -0,0 +1,478 @@ +/* + * Copyright (c) 2021 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 +#include + +#if defined(__ARM_FEATURE_SVE) && defined(__ARM_FP16_ARGS) + +namespace arm_conv { +namespace depthwise { + +void sve_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst_direct_impl( + const unsigned int n_tile_rows, + const unsigned int n_tile_cols, + const __fp16 *inptr, + int64_t ld_input_row, + int64_t ld_input_col, + __fp16 *outptr, + int64_t ld_output_row, + int64_t ld_output_col, + const void *params, + unsigned int n_channels, + const __fp16 activation_min, + const __fp16 activation_max +) +{ + struct Args + { + const uint64_t n_tile_rows, n_tile_cols; + const __fp16 *inptr; + const uint64_t ld_input_row; + const uint64_t ld_input_col; + __fp16 *outptr; + const uint64_t ld_output_row; + const uint64_t ld_output_col; + const void *params; + const __fp16 min, max; + + uint64_t tile_i = 0, tile_j = 0; + + Args( + const unsigned int n_tile_rows, + const unsigned int n_tile_cols, + const __fp16 *inptr, + int64_t ld_input_row, + int64_t ld_input_col, + __fp16 *outptr, + int64_t ld_output_row, + int64_t ld_output_col, + const void *params, + const float activation_min, + const float activation_max + ) : n_tile_rows(n_tile_rows), n_tile_cols(n_tile_cols), inptr(inptr), + ld_input_row(ld_input_row), ld_input_col(ld_input_col), outptr(outptr), + ld_output_row(ld_output_row), ld_output_col(ld_output_col), + params(params), min(activation_min), max(activation_max) + { + } + }; + + Args params_struct( + n_tile_rows, n_tile_cols, + inptr, ld_input_row, ld_input_col, + outptr, ld_output_row, ld_output_col, + params, activation_min, activation_max + ); + + __asm__ __volatile__( + "ptrue p3.b\n" + "mov x6, #0x0\n" + "mov x7, #0x0\n" + "1:" // Tile loop + "str x6, [%x[params_struct], %[offsetof_args_tile_i]]\n" + "mov x24, #0x3\n" + "str x7, [%x[params_struct], %[offsetof_args_tile_j]]\n" + "mov x23, #0x3\n" + "ldr x8, [%x[params_struct], %[offsetof_args_params]]\n" + "mov x17, #0x0\n" + "ldr x22, [%x[params_struct], %[offsetof_args_ld_input_row]]\n" + "cnth x16\n" + "ldr x15, [%x[params_struct], %[offsetof_args_ld_input_col]]\n" + "sub x21, XZR, x16\n" + "ldr x14, [%x[params_struct], %[offsetof_args_inptr]]\n" + "mul x19, x6, x22\n" // offset = tile_i * ld_input_row + "ldr x20, [%x[params_struct], %[offsetof_args_ld_output_row]]\n" + "madd x19, x7, x15, x19\n" // offset += tile_j * ld_input_col + "ldr x13, [%x[params_struct], %[offsetof_args_ld_output_col]]\n" + "mul x19, x19, x24\n" // offset *= kernel_stride * output_size + "ldr x12, [%x[params_struct], %[offsetof_args_outptr]]\n" + "add x14, x14, x19, LSL #1\n" // inptr[0] += offset * sizeof(__fp16) + "ld1rh { z18.h }, p3/Z, [%x[params_struct], %[offsetof_args_min]]\n" + "add x11, x14, x22, LSL #1\n" + "ld1rh { z17.h }, p3/Z, [%x[params_struct], %[offsetof_args_max]]\n" + "add x10, x11, x22, LSL #1\n" + "ld1h { z16.h }, p3/Z, [x8]\n" // Load from weights and bias + "mov z31.d, z16.d\n" + "ld1h { z0.h }, p3/Z, [x8, #1, MUL VL]\n" // Load from weights and bias + "add x9, x10, x22, LSL #1\n" + "mov z30.d, z16.d\n" + "ld1h { z1.h }, p3/Z, [x8, #2, MUL VL]\n" // Load from weights and bias + "add x28, x9, x22, LSL #1\n" + "mov z29.d, z16.d\n" + "ld1h { z2.h }, p3/Z, [x8, #3, MUL VL]\n" // Load from weights and bias + "add x27, x15, x15\n" + "mov z28.d, z16.d\n" + "ld1h { z3.h }, p3/Z, [x8, #4, MUL VL]\n" // Load from weights and bias + "add x26, x27, x15\n" + "mov z27.d, z16.d\n" + "ld1h { z4.h }, p3/Z, [x8, #5, MUL VL]\n" // Load from weights and bias + "add x25, x26, x15\n" + "mov z26.d, z16.d\n" + "ld1h { z5.h }, p3/Z, [x8, #6, MUL VL]\n" // Load from weights and bias + "mul x19, x6, x20\n" // offset = tile_i * ld_output_row + "mov z25.d, z16.d\n" + "ld1h { z6.h }, p3/Z, [x8, #7, MUL VL]\n" // Load from weights and bias + "madd x19, x7, x13, x19\n" // offset += tile_j * ld_output_col + "mov z24.d, z16.d\n" + "mul x19, x19, x23\n" // offset *= output_tile_size + "mov z23.d, z16.d\n" + "add x12, x12, x19, LSL #1\n" // outptrs[0] += offset * sizeof(__fp16) + "add x24, x13, x13\n" + "add x23, x12, x20, LSL #1\n" + "add x22, x23, x20, LSL #1\n" + "whilelt p2.h, XZR, %x[n_channels]\n" + "ld1h { z9.h }, p2/Z, [x10, x27, LSL #1]\n" // Load input point (2, 2) + "ld1h { z10.h }, p2/Z, [x14]\n" // Load input point (0, 0) + "addvl x8, x8, #16\n" + "ld1h { z11.h }, p2/Z, [x14, x25, LSL #1]\n" // Load input point (0, 4) + "cmp x16, %x[n_channels]\n" + "ld1h { z7.h }, p3/Z, [x8, #-8, MUL VL]\n" // Load from weights and bias + "ld1h { z8.h }, p3/Z, [x8, #-7, MUL VL]\n" // Load from weights and bias + "addvl x8, x8, #-6\n" + "ld1h { z12.h }, p2/Z, [x28]\n" // Load input point (4, 0) + "ld1h { z13.h }, p2/Z, [x11, x27, LSL #1]\n" // Load input point (1, 2) + "bge 3f\n" + "2:" // Tile loop: Channel loop + "fmla z31.h, p3/M, z8.h, z9.h\n" + "ld1h { z16.h }, p3/Z, [x8]\n" // Load from weights and bias + "whilelt p1.h, x16, %x[n_channels]\n" + "fmla z30.h, p3/M, z7.h, z9.h\n" + "inch x21\n" + "fmla z29.h, p3/M, z6.h, z9.h\n" + "mov p0.b, p2.b\n" + "fmla z28.h, p3/M, z5.h, z9.h\n" + "inch x17\n" + "fmla z27.h, p3/M, z4.h, z9.h\n" + "inch x16\n" + "fmla z26.h, p3/M, z3.h, z9.h\n" + "fmla z25.h, p3/M, z2.h, z9.h\n" + "fmla z24.h, p3/M, z1.h, z9.h\n" + "fmla z23.h, p3/M, z0.h, z9.h\n" + "fmla z31.h, p3/M, z0.h, z10.h\n" + "ld1h { z10.h }, p2/Z, [x10, x26, LSL #1]\n" // Load input point (2, 3) + "fmla z29.h, p3/M, z2.h, z11.h\n" + "ld1h { z11.h }, p2/Z, [x10, x15, LSL #1]\n" // Load input point (2, 1) + "fmla z25.h, p3/M, z6.h, z12.h\n" + "ld1h { z12.h }, p2/Z, [x28, x25, LSL #1]\n" // Load input point (4, 4) + "fmla z30.h, p3/M, z4.h, z13.h\n" + "fmla z31.h, p3/M, z5.h, z13.h\n" + "fmla z29.h, p3/M, z3.h, z13.h\n" + "fmla z28.h, p3/M, z2.h, z13.h\n" + "fmla z27.h, p3/M, z1.h, z13.h\n" + "fmla z26.h, p3/M, z0.h, z13.h\n" + "ld1h { z13.h }, p2/Z, [x14, x15, LSL #1]\n" // Load input point (0, 1) + "fmla z23.h, p3/M, z8.h, z12.h\n" + "ld1h { z12.h }, p2/Z, [x14, x26, LSL #1]\n" // Load input point (0, 3) + "fmla z31.h, p3/M, z7.h, z11.h\n" + "fmla z30.h, p3/M, z6.h, z11.h\n" + "fmla z28.h, p3/M, z4.h, z11.h\n" + "fmla z27.h, p3/M, z3.h, z11.h\n" + "fmla z25.h, p3/M, z1.h, z11.h\n" + "fmla z24.h, p3/M, z0.h, z11.h\n" + "ld1h { z11.h }, p2/Z, [x11]\n" // Load input point (1, 0) + "fmla z31.h, p3/M, z1.h, z13.h\n" + "fmla z30.h, p3/M, z0.h, z13.h\n" + "ld1h { z13.h }, p2/Z, [x11, x25, LSL #1]\n" // Load input point (1, 4) + "fmla z29.h, p3/M, z1.h, z12.h\n" + "fmla z27.h, p3/M, z5.h, z10.h\n" + "fmla z26.h, p3/M, z4.h, z10.h\n" + "fmla z30.h, p3/M, z2.h, z12.h\n" + "ld1h { z12.h }, p2/Z, [x9]\n" // Load input point (3, 0) + "fmla z29.h, p3/M, z7.h, z10.h\n" + "fmla z24.h, p3/M, z2.h, z10.h\n" + "fmla z23.h, p3/M, z1.h, z10.h\n" + "fmla z30.h, p3/M, z8.h, z10.h\n" + "ld1h { z10.h }, p2/Z, [x9, x27, LSL #1]\n" // Load input point (3, 2) + "fmla z31.h, p3/M, z3.h, z11.h\n" + "fmla z28.h, p3/M, z0.h, z11.h\n" + "ld1h { z11.h }, p2/Z, [x9, x25, LSL #1]\n" // Load input point (3, 4) + "fmla z29.h, p3/M, z5.h, z13.h\n" + "fmla z26.h, p3/M, z2.h, z13.h\n" + "ld1h { z13.h }, p2/Z, [x28, x15, LSL #1]\n" // Load input point (4, 1) + "fmla z25.h, p3/M, z3.h, z12.h\n" + "fmla z28.h, p3/M, z6.h, z12.h\n" + "ld1h { z12.h }, p2/Z, [x11, x15, LSL #1]\n" // Load input point (1, 1) + "fmla z27.h, p3/M, z7.h, z10.h\n" + "fmla z26.h, p3/M, z6.h, z10.h\n" + "fmla z25.h, p3/M, z5.h, z10.h\n" + "fmla z28.h, p3/M, z8.h, z10.h\n" + "fmla z24.h, p3/M, z4.h, z10.h\n" + "fmla z23.h, p3/M, z3.h, z10.h\n" + "fmla z26.h, p3/M, z8.h, z11.h\n" + "fmla z25.h, p3/M, z7.h, z13.h\n" + "fmla z24.h, p3/M, z6.h, z13.h\n" + "ld1h { z13.h }, p2/Z, [x28, x26, LSL #1]\n" // Load input point (4, 3) + "fmla z23.h, p3/M, z5.h, z11.h\n" + "ld1h { z11.h }, p2/Z, [x11, x26, LSL #1]\n" // Load input point (1, 3) + "addvl x11, x11, #1\n" + "fmla z31.h, p3/M, z4.h, z12.h\n" + "fmla z30.h, p3/M, z3.h, z12.h\n" + "fmla z28.h, p3/M, z1.h, z12.h\n" + "fmla z27.h, p3/M, z0.h, z12.h\n" + "ld1h { z12.h }, p2/Z, [x9, x15, LSL #1]\n" // Load input point (3, 1) + "fmla z29.h, p3/M, z4.h, z11.h\n" + "fmla z30.h, p3/M, z5.h, z11.h\n" + "fmla z26.h, p3/M, z1.h, z11.h\n" + "fmla z27.h, p3/M, z2.h, z11.h\n" + "ld1h { z11.h }, p2/Z, [x14, x27, LSL #1]\n" // Load input point (0, 2) + "addvl x14, x14, #1\n" + "fmla z24.h, p3/M, z8.h, z13.h\n" + "ld1h { z10.h }, p1/Z, [x14]\n" // Load input point (0, 0) + "fmla z23.h, p3/M, z7.h, z13.h\n" + "ld1h { z13.h }, p2/Z, [x9, x26, LSL #1]\n" // Load input point (3, 3) + "addvl x9, x9, #1\n" + "fmla z28.h, p3/M, z7.h, z12.h\n" + "fmla z27.h, p3/M, z6.h, z12.h\n" + "fmla z25.h, p3/M, z4.h, z12.h\n" + "fmla z24.h, p3/M, z3.h, z12.h\n" + "ld1h { z12.h }, p2/Z, [x10]\n" // Load input point (2, 0) + "fmla z31.h, p3/M, z2.h, z11.h\n" + "fmla z30.h, p3/M, z1.h, z11.h\n" + "ld1h { z1.h }, p3/Z, [x8, #2, MUL VL]\n" // Load from weights and bias + "fmla z29.h, p3/M, z0.h, z11.h\n" + "ld1h { z11.h }, p2/Z, [x10, x25, LSL #1]\n" // Load input point (2, 4) + "addvl x10, x10, #1\n" + "fmla z27.h, p3/M, z8.h, z13.h\n" + "ld1h { z9.h }, p1/Z, [x10, x27, LSL #1]\n" // Load input point (2, 2) + "fmla z26.h, p3/M, z7.h, z13.h\n" + "fmla z24.h, p3/M, z5.h, z13.h\n" + "fmla z23.h, p3/M, z4.h, z13.h\n" + "ld1h { z13.h }, p2/Z, [x28, x27, LSL #1]\n" // Load input point (4, 2) + "whilelt p2.h, x17, %x[n_channels]\n" + "fmla z31.h, p3/M, z6.h, z12.h\n" + "ld1h { z4.h }, p3/Z, [x8, #5, MUL VL]\n" // Load from weights and bias + "addvl x28, x28, #1\n" + "fmla z28.h, p3/M, z3.h, z12.h\n" + "ld1h { z3.h }, p3/Z, [x8, #4, MUL VL]\n" // Load from weights and bias + "cmp x16, %x[n_channels]\n" + "fmla z25.h, p3/M, z0.h, z12.h\n" + "ld1h { z12.h }, p1/Z, [x28]\n" // Load input point (4, 0) + "fmla z29.h, p3/M, z8.h, z11.h\n" + "ld1h { z0.h }, p3/Z, [x8, #1, MUL VL]\n" // Load from weights and bias + "fmla z26.h, p3/M, z5.h, z11.h\n" + "ld1h { z5.h }, p3/Z, [x8, #6, MUL VL]\n" // Load from weights and bias + "fmla z23.h, p3/M, z2.h, z11.h\n" + "ld1h { z11.h }, p1/Z, [x14, x25, LSL #1]\n" // Load input point (0, 4) + "fmla z24.h, p3/M, z7.h, z13.h\n" + "ld1h { z2.h }, p3/Z, [x8, #3, MUL VL]\n" // Load from weights and bias + "fmla z25.h, p3/M, z8.h, z13.h\n" + "fmax z31.h, p3/M, z31.h, z18.h\n" + "fmla z23.h, p3/M, z6.h, z13.h\n" + "ld1h { z13.h }, p1/Z, [x11, x27, LSL #1]\n" // Load input point (1, 2) + "fmax z30.h, p3/M, z30.h, z18.h\n" + "ld1h { z6.h }, p3/Z, [x8, #7, MUL VL]\n" // Load from weights and bias + "addvl x8, x8, #16\n" + "fmin z31.h, p3/M, z31.h, z17.h\n" + "ld1h { z7.h }, p3/Z, [x8, #-8, MUL VL]\n" // Load from weights and bias + "fmax z29.h, p3/M, z29.h, z18.h\n" + "ld1h { z8.h }, p3/Z, [x8, #-7, MUL VL]\n" // Load from weights and bias + "addvl x8, x8, #-6\n" + "fmin z30.h, p3/M, z30.h, z17.h\n" + "st1h { z31.h }, p0, [x12]\n" // Store output point (0, 0) + "mov z31.d, z16.d\n" + "fmin z29.h, p3/M, z29.h, z17.h\n" + "st1h { z30.h }, p0, [x12, x13, LSL #1]\n" // Store output point (0, 1) + "mov z30.d, z16.d\n" + "fmax z28.h, p3/M, z28.h, z18.h\n" + "st1h { z29.h }, p0, [x12, x24, LSL #1]\n" // Store output point (0, 2) + "mov z29.d, z16.d\n" + "addvl x12, x12, #1\n" + "fmax z27.h, p3/M, z27.h, z18.h\n" + "fmax z26.h, p3/M, z26.h, z18.h\n" + "fmin z28.h, p3/M, z28.h, z17.h\n" + "st1h { z28.h }, p0, [x23]\n" // Store output point (1, 0) + "mov z28.d, z16.d\n" + "fmin z27.h, p3/M, z27.h, z17.h\n" + "st1h { z27.h }, p0, [x23, x13, LSL #1]\n" // Store output point (1, 1) + "mov z27.d, z16.d\n" + "fmin z26.h, p3/M, z26.h, z17.h\n" + "st1h { z26.h }, p0, [x23, x24, LSL #1]\n" // Store output point (1, 2) + "mov z26.d, z16.d\n" + "addvl x23, x23, #1\n" + "fmax z25.h, p3/M, z25.h, z18.h\n" + "fmax z24.h, p3/M, z24.h, z18.h\n" + "fmax z23.h, p3/M, z23.h, z18.h\n" + "fmin z25.h, p3/M, z25.h, z17.h\n" + "st1h { z25.h }, p0, [x22]\n" // Store output point (2, 0) + "mov z25.d, z16.d\n" + "fmin z24.h, p3/M, z24.h, z17.h\n" + "st1h { z24.h }, p0, [x22, x13, LSL #1]\n" // Store output point (2, 1) + "mov z24.d, z16.d\n" + "fmin z23.h, p3/M, z23.h, z17.h\n" + "st1h { z23.h }, p0, [x22, x24, LSL #1]\n" // Store output point (2, 2) + "mov z23.d, z16.d\n" + "addvl x22, x22, #1\n" + "blt 2b\n" + "3:" // Tile loop: Channel tail + "fmla z31.h, p3/M, z8.h, z9.h\n" + "ldr x6, [%x[params_struct], %[offsetof_args_tile_i]]\n" + "mov p0.b, p2.b\n" + "fmla z30.h, p3/M, z7.h, z9.h\n" + "ldr x7, [%x[params_struct], %[offsetof_args_tile_j]]\n" + "add x21, x6, #0x1\n" + "fmla z29.h, p3/M, z6.h, z9.h\n" + "ldr x20, [%x[params_struct], %[offsetof_args_n_tile_rows]]\n" + "fmla z28.h, p3/M, z5.h, z9.h\n" + "ldr x19, [%x[params_struct], %[offsetof_args_n_tile_cols]]\n" + "add x7, x7, #0x1\n" + "fmla z27.h, p3/M, z4.h, z9.h\n" + "cmp x7, x19\n" + "fmla z26.h, p3/M, z3.h, z9.h\n" + "fmla z25.h, p3/M, z2.h, z9.h\n" + "csel x7, x7, XZR, LT\n" + "fmla z24.h, p3/M, z1.h, z9.h\n" + "csel x6, x6, x21, LT\n" + "fmla z23.h, p3/M, z0.h, z9.h\n" + "cmp x6, x20\n" + "fmla z31.h, p3/M, z0.h, z10.h\n" + "ld1h { z10.h }, p2/Z, [x10, x26, LSL #1]\n" // Load input point (2, 3) + "fmla z29.h, p3/M, z2.h, z11.h\n" + "ld1h { z11.h }, p2/Z, [x10, x15, LSL #1]\n" // Load input point (2, 1) + "fmla z25.h, p3/M, z6.h, z12.h\n" + "ld1h { z12.h }, p2/Z, [x28, x25, LSL #1]\n" // Load input point (4, 4) + "fmla z30.h, p3/M, z4.h, z13.h\n" + "fmla z31.h, p3/M, z5.h, z13.h\n" + "fmla z29.h, p3/M, z3.h, z13.h\n" + "fmla z28.h, p3/M, z2.h, z13.h\n" + "fmla z27.h, p3/M, z1.h, z13.h\n" + "fmla z26.h, p3/M, z0.h, z13.h\n" + "ld1h { z13.h }, p2/Z, [x14, x15, LSL #1]\n" // Load input point (0, 1) + "fmla z23.h, p3/M, z8.h, z12.h\n" + "ld1h { z12.h }, p2/Z, [x14, x26, LSL #1]\n" // Load input point (0, 3) + "fmla z31.h, p3/M, z7.h, z11.h\n" + "fmla z30.h, p3/M, z6.h, z11.h\n" + "fmla z28.h, p3/M, z4.h, z11.h\n" + "fmla z27.h, p3/M, z3.h, z11.h\n" + "fmla z25.h, p3/M, z1.h, z11.h\n" + "fmla z24.h, p3/M, z0.h, z11.h\n" + "ld1h { z11.h }, p2/Z, [x11]\n" // Load input point (1, 0) + "fmla z31.h, p3/M, z1.h, z13.h\n" + "fmla z30.h, p3/M, z0.h, z13.h\n" + "ld1h { z13.h }, p2/Z, [x11, x25, LSL #1]\n" // Load input point (1, 4) + "fmla z29.h, p3/M, z1.h, z12.h\n" + "fmla z27.h, p3/M, z5.h, z10.h\n" + "fmla z26.h, p3/M, z4.h, z10.h\n" + "fmla z30.h, p3/M, z2.h, z12.h\n" + "ld1h { z12.h }, p2/Z, [x9]\n" // Load input point (3, 0) + "fmla z29.h, p3/M, z7.h, z10.h\n" + "fmla z24.h, p3/M, z2.h, z10.h\n" + "fmla z23.h, p3/M, z1.h, z10.h\n" + "fmla z30.h, p3/M, z8.h, z10.h\n" + "ld1h { z10.h }, p2/Z, [x9, x27, LSL #1]\n" // Load input point (3, 2) + "fmla z31.h, p3/M, z3.h, z11.h\n" + "fmla z28.h, p3/M, z0.h, z11.h\n" + "ld1h { z11.h }, p2/Z, [x9, x25, LSL #1]\n" // Load input point (3, 4) + "fmla z29.h, p3/M, z5.h, z13.h\n" + "fmla z26.h, p3/M, z2.h, z13.h\n" + "ld1h { z13.h }, p2/Z, [x28, x15, LSL #1]\n" // Load input point (4, 1) + "fmla z25.h, p3/M, z3.h, z12.h\n" + "fmla z28.h, p3/M, z6.h, z12.h\n" + "ld1h { z12.h }, p2/Z, [x11, x15, LSL #1]\n" // Load input point (1, 1) + "fmla z27.h, p3/M, z7.h, z10.h\n" + "fmla z26.h, p3/M, z6.h, z10.h\n" + "fmla z25.h, p3/M, z5.h, z10.h\n" + "fmla z28.h, p3/M, z8.h, z10.h\n" + "fmla z24.h, p3/M, z4.h, z10.h\n" + "fmla z23.h, p3/M, z3.h, z10.h\n" + "fmla z26.h, p3/M, z8.h, z11.h\n" + "fmla z25.h, p3/M, z7.h, z13.h\n" + "fmla z24.h, p3/M, z6.h, z13.h\n" + "ld1h { z13.h }, p2/Z, [x28, x26, LSL #1]\n" // Load input point (4, 3) + "fmla z23.h, p3/M, z5.h, z11.h\n" + "ld1h { z11.h }, p2/Z, [x11, x26, LSL #1]\n" // Load input point (1, 3) + "fmla z31.h, p3/M, z4.h, z12.h\n" + "fmla z30.h, p3/M, z3.h, z12.h\n" + "fmla z28.h, p3/M, z1.h, z12.h\n" + "fmla z27.h, p3/M, z0.h, z12.h\n" + "ld1h { z12.h }, p2/Z, [x9, x15, LSL #1]\n" // Load input point (3, 1) + "fmla z29.h, p3/M, z4.h, z11.h\n" + "fmla z30.h, p3/M, z5.h, z11.h\n" + "fmla z26.h, p3/M, z1.h, z11.h\n" + "fmla z27.h, p3/M, z2.h, z11.h\n" + "ld1h { z11.h }, p2/Z, [x14, x27, LSL #1]\n" // Load input point (0, 2) + "fmla z24.h, p3/M, z8.h, z13.h\n" + "fmla z23.h, p3/M, z7.h, z13.h\n" + "ld1h { z13.h }, p2/Z, [x9, x26, LSL #1]\n" // Load input point (3, 3) + "fmla z28.h, p3/M, z7.h, z12.h\n" + "fmla z27.h, p3/M, z6.h, z12.h\n" + "fmla z25.h, p3/M, z4.h, z12.h\n" + "fmla z24.h, p3/M, z3.h, z12.h\n" + "ld1h { z12.h }, p2/Z, [x10]\n" // Load input point (2, 0) + "fmla z31.h, p3/M, z2.h, z11.h\n" + "fmla z30.h, p3/M, z1.h, z11.h\n" + "fmla z29.h, p3/M, z0.h, z11.h\n" + "ld1h { z11.h }, p2/Z, [x10, x25, LSL #1]\n" // Load input point (2, 4) + "fmla z27.h, p3/M, z8.h, z13.h\n" + "fmla z26.h, p3/M, z7.h, z13.h\n" + "fmla z24.h, p3/M, z5.h, z13.h\n" + "fmla z23.h, p3/M, z4.h, z13.h\n" + "ld1h { z13.h }, p2/Z, [x28, x27, LSL #1]\n" // Load input point (4, 2) + "fmla z31.h, p3/M, z6.h, z12.h\n" + "fmla z28.h, p3/M, z3.h, z12.h\n" + "fmla z25.h, p3/M, z0.h, z12.h\n" + "fmla z29.h, p3/M, z8.h, z11.h\n" + "fmla z26.h, p3/M, z5.h, z11.h\n" + "fmla z23.h, p3/M, z2.h, z11.h\n" + "fmla z25.h, p3/M, z8.h, z13.h\n" + "fmla z24.h, p3/M, z7.h, z13.h\n" + "fmax z31.h, p3/M, z31.h, z18.h\n" + "fmla z23.h, p3/M, z6.h, z13.h\n" + "fmax z30.h, p3/M, z30.h, z18.h\n" + "fmax z29.h, p3/M, z29.h, z18.h\n" + "fmin z31.h, p3/M, z31.h, z17.h\n" + "st1h { z31.h }, p0, [x12]\n" // Store output point (0, 0) + "fmin z30.h, p3/M, z30.h, z17.h\n" + "fmin z29.h, p3/M, z29.h, z17.h\n" + "st1h { z30.h }, p0, [x12, x13, LSL #1]\n" // Store output point (0, 1) + "fmax z28.h, p3/M, z28.h, z18.h\n" + "fmax z27.h, p3/M, z27.h, z18.h\n" + "st1h { z29.h }, p0, [x12, x24, LSL #1]\n" // Store output point (0, 2) + "fmax z26.h, p3/M, z26.h, z18.h\n" + "fmax z25.h, p3/M, z25.h, z18.h\n" + "fmax z24.h, p3/M, z24.h, z18.h\n" + "fmin z28.h, p3/M, z28.h, z17.h\n" + "st1h { z28.h }, p0, [x23]\n" // Store output point (1, 0) + "fmin z27.h, p3/M, z27.h, z17.h\n" + "fmin z26.h, p3/M, z26.h, z17.h\n" + "st1h { z27.h }, p0, [x23, x13, LSL #1]\n" // Store output point (1, 1) + "fmin z25.h, p3/M, z25.h, z17.h\n" + "fmin z24.h, p3/M, z24.h, z17.h\n" + "st1h { z26.h }, p0, [x23, x24, LSL #1]\n" // Store output point (1, 2) + "fmax z23.h, p3/M, z23.h, z18.h\n" + "st1h { z25.h }, p0, [x22]\n" // Store output point (2, 0) + "fmin z23.h, p3/M, z23.h, z17.h\n" + "st1h { z24.h }, p0, [x22, x13, LSL #1]\n" // Store output point (2, 1) + "st1h { z23.h }, p0, [x22, x24, LSL #1]\n" // Store output point (2, 2) + "blt 1b\n" + : + : [n_channels] "r" ((unsigned long) n_channels), [offsetof_args_inptr] "I" (offsetof(Args, inptr)), [offsetof_args_ld_input_col] "I" (offsetof(Args, ld_input_col)), [offsetof_args_ld_input_row] "I" (offsetof(Args, ld_input_row)), [offsetof_args_ld_output_col] "I" (offsetof(Args, ld_output_col)), [offsetof_args_ld_output_row] "I" (offsetof(Args, ld_output_row)), [offsetof_args_max] "I" (offsetof(Args, max)), [offsetof_args_min] "I" (offsetof(Args, min)), [offsetof_args_n_tile_cols] "I" (offsetof(Args, n_tile_cols)), [offsetof_args_n_tile_rows] "I" (offsetof(Args, n_tile_rows)), [offsetof_args_outptr] "I" (offsetof(Args, outptr)), [offsetof_args_params] "I" (offsetof(Args, params)), [offsetof_args_tile_i] "I" (offsetof(Args, tile_i)), [offsetof_args_tile_j] "I" (offsetof(Args, tile_j)), [params_struct] "r" (¶ms_struct) + : "cc", "memory", "p0", "p1", "p2", "p3", "x6", "x7", "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z16", "z17", "z18", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31" + ); +} + +} // namespace depthwise +} // namespace arm_conv + +#endif // defined(__ARM_FEATURE_SVE) && defined(__ARM_FP16_ARGS) -- cgit v1.2.1