| // Auto-generated file. Do not edit! |
| // Template: src/f32-dwconv2d-chw/3x3p1-wasmsimd-1x4-acc3.c.in |
| // Generator: tools/xngen |
| // |
| // Copyright 2020 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 <assert.h> |
| |
| #include <wasm_simd128.h> |
| |
| |
| #include <xnnpack/dwconv.h> |
| #include <xnnpack/math.h> |
| |
| static v128_t concat_hi_f32(v128_t a, v128_t b) { |
| return wasm_v32x4_shuffle(a, b, 2, 3, 4+2, 4+3); |
| } |
| |
| static v128_t rotright_f32(v128_t a) { |
| return wasm_v32x4_shuffle(a, a, 3, 0, 1, 2); |
| } |
| |
| static v128_t rotleft_f32(v128_t a) { |
| return wasm_v32x4_shuffle(a, a, 1, 2, 3, 0); |
| } |
| |
| static v128_t movess_f32(v128_t a, v128_t b) { |
| return wasm_v32x4_shuffle(a, b, 4, 1, 2, 3); |
| } |
| |
| |
| void xnn_f32_dwconv2d_chw_ukernel_3x3p1__wasmsimd_x86_1x4_acc3( |
| size_t input_height, |
| size_t input_width, |
| const float* input, |
| const float* weights, |
| const float* zero, |
| float* output, |
| uint32_t padding_top, |
| const union xnn_f32_chw_params params[restrict XNN_MIN_ELEMENTS(1)]) |
| { |
| assert(input_height != 0); |
| assert(input_width != 0); |
| assert(input_width % sizeof(float) == 0); |
| assert(padding_top == 1); |
| |
| const v128_t vmask = wasm_v128_load(params->scalar.mask); |
| const v128_t vmax = wasm_v32x4_load_splat(¶ms->scalar.max); |
| const v128_t vmin = wasm_v32x4_load_splat(¶ms->scalar.min); |
| |
| const v128_t vbias = wasm_v32x4_load_splat(weights); |
| const v128_t vk00 = wasm_v32x4_load_splat(weights + 1); |
| const v128_t vk01 = wasm_v32x4_load_splat(weights + 2); |
| const v128_t vk02 = wasm_v32x4_load_splat(weights + 3); |
| const v128_t vk10 = wasm_v32x4_load_splat(weights + 4); |
| const v128_t vk11 = wasm_v32x4_load_splat(weights + 5); |
| const v128_t vk12 = wasm_v32x4_load_splat(weights + 6); |
| const v128_t vk20 = wasm_v32x4_load_splat(weights + 7); |
| const v128_t vk21 = wasm_v32x4_load_splat(weights + 8); |
| const v128_t vk22 = wasm_v32x4_load_splat(weights + 9); |
| |
| const size_t input_decrement = round_up_po2(input_width, 4 * sizeof(float)); |
| |
| const float* i0 = zero; |
| const float* i1 = input; |
| const float* i2 = (const float*) ((uintptr_t) i1 + input_width); |
| |
| size_t output_height = input_height; |
| do { |
| if XNN_UNPREDICTABLE(output_height == 1) { |
| i2 = zero; |
| } |
| |
| // vi0x3012 = ( vi02, vi01, vi00, vi03 ) |
| v128_t vi0x3012 = wasm_f32x4_splat(0.0f); |
| // vi1x3012 = ( vi12, vi11, vi10, vi13 ) |
| v128_t vi1x3012 = wasm_f32x4_splat(0.0f); |
| // vi2x3012 = ( vi22, vi21, vi20, vi13 ) |
| v128_t vi2x3012 = wasm_f32x4_splat(0.0f); |
| // vi0x4567 = ( vi07, vi06, vi05, vi04 ) |
| v128_t vi0x4567 = wasm_v128_load(i0); |
| i0 += 4; |
| // vi1x4567 = ( vi17, vi16, vi15, vi14 ) |
| v128_t vi1x4567 = wasm_v128_load(i1); |
| i1 += 4; |
| // vi2x4567 = ( vi27, vi26, vi25, vi24 ) |
| v128_t vi2x4567 = wasm_v128_load(i2); |
| i2 += 4; |
| |
| size_t w = input_width; |
| for (; w > 4 * sizeof(float); w -= 4 * sizeof(float)) { |
| v128_t vo4567p0 = vbias; |
| |
| // vi0x89AB = ( vi0B, vi0A, vi09, vi08 ) |
| const v128_t vi0x89AB = wasm_v128_load(i0); |
| i0 += 4; |
| // vi1x89AB = ( vi1B, vi0A, vi09, vi08 ) |
| const v128_t vi1x89AB = wasm_v128_load(i1); |
| i1 += 4; |
| // vi2x89AB = ( vi2B, vi0A, vi09, vi08 ) |
| const v128_t vi2x89AB = wasm_v128_load(i2); |
| i2 += 4; |
| |
| // vi0x7456 = ( vi06, vi05, vi04, vi07 ) |
| const v128_t vi0x7456 = rotright_f32(vi0x4567); |
| // vi1x7456 = ( vi16, vi15, vi14, vi17 ) |
| const v128_t vi1x7456 = rotright_f32(vi1x4567); |
| // vi2x7456 = ( vi26, vi25, vi24, vi27 ) |
| const v128_t vi2x7456 = rotright_f32(vi2x4567); |
| |
| vo4567p0 = wasm_f32x4_add(vo4567p0, wasm_f32x4_mul(vi0x4567, vk01)); |
| v128_t vo4567p1 = wasm_f32x4_mul(vi1x4567, vk11); |
| v128_t vo4567p2 = wasm_f32x4_mul(vi2x4567, vk21); |
| |
| // vi0x3456 = ( vi06, vi05, vi04, vi03 ) |
| const v128_t vi0x3456 = movess_f32(vi0x7456, vi0x3012); |
| // vi1x3456 = ( vi16, vi15, vi14, vi13 ) |
| const v128_t vi1x3456 = movess_f32(vi1x7456, vi1x3012); |
| // vi2x3456 = ( vi26, vi25, vi24, vi23 ) |
| const v128_t vi2x3456 = movess_f32(vi2x7456, vi2x3012); |
| |
| vo4567p0 = wasm_f32x4_add(vo4567p0, wasm_f32x4_mul(vi0x3456, vk00)); |
| vo4567p1 = wasm_f32x4_add(vo4567p1, wasm_f32x4_mul(vi1x3456, vk10)); |
| vo4567p2 = wasm_f32x4_add(vo4567p2, wasm_f32x4_mul(vi2x3456, vk20)); |
| |
| vi0x3012 = vi0x7456; |
| vi1x3012 = vi1x7456; |
| vi2x3012 = vi2x7456; |
| |
| // vi0x8567 = ( vi07, vi06, vi05, vi08 ) |
| const v128_t vi0x8567 = movess_f32(vi0x4567, vi0x89AB); |
| // vi1x8567 = ( vi17, vi16, vi15, vi18 ) |
| const v128_t vi1x8567 = movess_f32(vi1x4567, vi1x89AB); |
| // vi2x8567 = ( vi27, vi26, vi25, vi28 ) |
| const v128_t vi2x8567 = movess_f32(vi2x4567, vi2x89AB); |
| |
| // vi0x5678 = ( vi08, vi07, vi06, vi05 ) |
| const v128_t vi0x5678 = rotleft_f32(vi0x8567); |
| // vi1x5678 = ( vi18, vi17, vi16, vi15 ) |
| const v128_t vi1x5678 = rotleft_f32(vi1x8567); |
| // vi2x5678 = ( vi28, vi27, vi26, vi25 ) |
| const v128_t vi2x5678 = rotleft_f32(vi2x8567); |
| |
| vo4567p0 = wasm_f32x4_add(vo4567p0, wasm_f32x4_mul(vi0x5678, vk02)); |
| vo4567p1 = wasm_f32x4_add(vo4567p1, wasm_f32x4_mul(vi1x5678, vk12)); |
| vo4567p2 = wasm_f32x4_add(vo4567p2, wasm_f32x4_mul(vi2x5678, vk22)); |
| |
| vi0x4567 = vi0x89AB; |
| vi1x4567 = vi1x89AB; |
| vi2x4567 = vi2x89AB; |
| |
| v128_t vo = wasm_f32x4_add(vo4567p0, vo4567p1); |
| vo = wasm_f32x4_add(vo, vo4567p2); |
| |
| vo = wasm_v128_bitselect(vmin, vo, wasm_f32x4_lt(vo, vmin)); |
| vo = wasm_v128_bitselect(vo, vmax, wasm_f32x4_le(vo, vmax)); |
| |
| wasm_v128_store(output, vo); |
| output += 4; |
| } |
| // Always process the last block of 1..4 pixels. |
| assert(w >= 1 * sizeof(float)); |
| assert(w <= 4 * sizeof(float)); |
| { |
| v128_t vo4567p0 = vbias; |
| |
| vi0x4567 = wasm_v128_and(vmask, vi0x4567); |
| vi1x4567 = wasm_v128_and(vmask, vi1x4567); |
| vi2x4567 = wasm_v128_and(vmask, vi2x4567); |
| |
| // vi0x7456 = ( vi06, vi05, vi04, vi07 ) |
| const v128_t vi0x7456 = rotright_f32(vi0x4567); |
| // vi1x7456 = ( vi16, vi15, vi14, vi17 ) |
| const v128_t vi1x7456 = rotright_f32(vi1x4567); |
| // vi2x7456 = ( vi26, vi25, vi24, vi27 ) |
| const v128_t vi2x7456 = rotright_f32(vi2x4567); |
| |
| vo4567p0 = wasm_f32x4_add(vo4567p0, wasm_f32x4_mul(vi0x4567, vk01)); |
| v128_t vo4567p1 = wasm_f32x4_mul(vi1x4567, vk11); |
| v128_t vo4567p2 = wasm_f32x4_mul(vi2x4567, vk21); |
| |
| // vi0x3456 = ( vi06, vi05, vi04, vi03 ) |
| const v128_t vi0x3456 = movess_f32(vi0x7456, vi0x3012); |
| // vi1x3456 = ( vi16, vi15, vi14, vi13 ) |
| const v128_t vi1x3456 = movess_f32(vi1x7456, vi1x3012); |
| // vi2x3456 = ( vi26, vi25, vi24, vi23 ) |
| const v128_t vi2x3456 = movess_f32(vi2x7456, vi2x3012); |
| |
| vo4567p0 = wasm_f32x4_add(vo4567p0, wasm_f32x4_mul(vi0x3456, vk00)); |
| vo4567p1 = wasm_f32x4_add(vo4567p1, wasm_f32x4_mul(vi1x3456, vk10)); |
| vo4567p2 = wasm_f32x4_add(vo4567p2, wasm_f32x4_mul(vi2x3456, vk20)); |
| |
| const v128_t vzero = wasm_f32x4_splat(0.0f); |
| // vi0x8567 = ( vi07, vi06, vi05, 0.0 ) |
| const v128_t vi0x8567 = movess_f32(vi0x4567, vzero); |
| // vi1x8567 = ( vi17, vi16, vi15, 0.0 ) |
| const v128_t vi1x8567 = movess_f32(vi1x4567, vzero); |
| // vi2x8567 = ( vi27, vi26, vi25, 0.0 ) |
| const v128_t vi2x8567 = movess_f32(vi2x4567, vzero); |
| |
| // vi0x5678 = ( vi08, vi07, vi06, vi05 ) |
| const v128_t vi0x5678 = rotleft_f32(vi0x8567); |
| // vi1x5678 = ( vi18, vi17, vi16, vi15 ) |
| const v128_t vi1x5678 = rotleft_f32(vi1x8567); |
| // vi2x5678 = ( vi28, vi27, vi26, vi25 ) |
| const v128_t vi2x5678 = rotleft_f32(vi2x8567); |
| |
| vo4567p0 = wasm_f32x4_add(vo4567p0, wasm_f32x4_mul(vi0x5678, vk02)); |
| vo4567p1 = wasm_f32x4_add(vo4567p1, wasm_f32x4_mul(vi1x5678, vk12)); |
| vo4567p2 = wasm_f32x4_add(vo4567p2, wasm_f32x4_mul(vi2x5678, vk22)); |
| |
| v128_t vo = wasm_f32x4_add(vo4567p0, vo4567p1); |
| vo = wasm_f32x4_add(vo, vo4567p2); |
| |
| vo = wasm_v128_bitselect(vmin, vo, wasm_f32x4_lt(vo, vmin)); |
| vo = wasm_v128_bitselect(vo, vmax, wasm_f32x4_le(vo, vmax)); |
| |
| if XNN_LIKELY(w == 4 * sizeof(float)) { |
| wasm_v128_store(output, vo); |
| output += 4; |
| } else { |
| if (w & (2 * sizeof(float))) { |
| *((double*) output) = wasm_f64x2_extract_lane(vo, 0); |
| output += 2; |
| vo = concat_hi_f32(vo, vo); |
| } |
| if (w & (1 * sizeof(float))) { |
| *output = wasm_f32x4_extract_lane(vo, 0); |
| output += 1; |
| } |
| } |
| } |
| |
| i0 = (const float*) ((uintptr_t) i1 - input_decrement); |
| i1 = (const float*) ((uintptr_t) i2 - input_decrement); |
| i2 = (const float*) ((uintptr_t) i1 + input_width); |
| } while (--output_height != 0); |
| } |