src/f32-dwconv2d-chw/3x3s2p1-wasmsimd-1x4-acc3.c.in - platform/external/XNNPACK - Gitiles

 // 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_even_f32(v128_t a, v128_t b) {
   return wasm_v32x4_shuffle(a, b, 0, 2, 4+0, 4+2);
 }

 static v128_t concat_odd_f32(v128_t a, v128_t b) {
   return wasm_v32x4_shuffle(a, b, 1, 3, 4+1, 4+3);
 }

 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 movess_f32(v128_t a, v128_t b) {
   return wasm_v32x4_shuffle(a, b, 4, 1, 2, 3);
 }

 $ARCH_SUFFIX = "_x86" if X86 else "_arm"

 void xnn_f32_dwconv2d_chw_ukernel_3x3s2p1__wasmsimd${ARCH_SUFFIX}_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 >= 0);
   assert(padding_top <= 1);

   const v128_t vmask_even = wasm_v128_load(params->scalar.mask_even);
   const v128_t vmask_odd  = wasm_v128_load(params->scalar.mask_odd);
   const v128_t vmax = wasm_v32x4_load_splat(&params->scalar.max);
   const v128_t vmin = wasm_v32x4_load_splat(&params->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_down_po2(input_width, 4 /* SIMD output width */ * 2 /* subsampling */ * sizeof(float));

   const float* i0 = (const float*) ((uintptr_t) input - ((-padding_top) & input_width));
   const float* i1 = (const float*) ((uintptr_t) i0 + input_width);
   if XNN_UNPREDICTABLE(padding_top != 0) {
     i0 = zero;
   }
   const float* i2 = (const float*) ((uintptr_t) i1 + input_width);

   size_t padded_input_height = input_height + padding_top + 1 /* padding bottom */;
   size_t output_height = (padded_input_height - 3 /* kernel size */ + 2 /* subsampling */) / 2;
   do {
     if XNN_UNPREDICTABLE(padded_input_height <= 3) {
       i2 = zero;
     }

     v128_t vi0x7531 = wasm_f32x4_splat(0.0f);
     v128_t vi1x7531 = wasm_f32x4_splat(0.0f);
     v128_t vi2x7531 = wasm_f32x4_splat(0.0f);

     size_t w = input_width;
     for (; w >= 8 * sizeof(float); w -= 8 * sizeof(float)) {
       v128_t vo8ACEp0 = vbias;

       const v128_t vi0x89AB = wasm_v128_load(i0);
       const v128_t vi1x89AB = wasm_v128_load(i1);
       const v128_t vi2x89AB = wasm_v128_load(i2);

       const v128_t vi0xCDEF = wasm_v128_load(i0 + 4);
       i0 += 8;
       const v128_t vi1xCDEF = wasm_v128_load(i1 + 4);
       i1 += 8;
       const v128_t vi2xCDEF = wasm_v128_load(i2 + 4);
       i2 += 8;

       const v128_t vi0x8ACE = concat_even_f32(vi0x89AB, vi0xCDEF);
       const v128_t vi0x9BDF = concat_odd_f32(vi0x89AB, vi0xCDEF);
       const v128_t vi1x8ACE = concat_even_f32(vi1x89AB, vi1xCDEF);
       const v128_t vi1x9BDF = concat_odd_f32(vi1x89AB, vi1xCDEF);
       const v128_t vi2x8ACE = concat_even_f32(vi2x89AB, vi2xCDEF);
       const v128_t vi2x9BDF = concat_odd_f32(vi2x89AB, vi2xCDEF);

       vo8ACEp0 = wasm_f32x4_add(vo8ACEp0, wasm_f32x4_mul(vi0x8ACE, vk01));
       v128_t vo8ACEp1 = wasm_f32x4_mul(vi1x8ACE, vk11);
       v128_t vo8ACEp2 = wasm_f32x4_mul(vi2x8ACE, vk21);

       const v128_t vi0xF9BD = rotright_f32(vi0x9BDF);
       const v128_t vi1xF9BD = rotright_f32(vi1x9BDF);
       const v128_t vi2xF9BD = rotright_f32(vi2x9BDF);

       vo8ACEp0 = wasm_f32x4_add(vo8ACEp0, wasm_f32x4_mul(vi0x9BDF, vk02));
       vo8ACEp1 = wasm_f32x4_add(vo8ACEp1, wasm_f32x4_mul(vi1x9BDF, vk12));
       vo8ACEp2 = wasm_f32x4_add(vo8ACEp2, wasm_f32x4_mul(vi2x9BDF, vk22));

       const v128_t vi0x7BDF = movess_f32(vi0xF9BD, vi0x7531);
       const v128_t vi1x7BDF = movess_f32(vi1xF9BD, vi1x7531);
       const v128_t vi2x7BDF = movess_f32(vi2xF9BD, vi2x7531);

       vi0x7531 = vi0xF9BD;
       vi1x7531 = vi1xF9BD;
       vi2x7531 = vi2xF9BD;

       vo8ACEp0 = wasm_f32x4_add(vo8ACEp0, wasm_f32x4_mul(vi0x7BDF, vk00));
       vo8ACEp1 = wasm_f32x4_add(vo8ACEp1, wasm_f32x4_mul(vi1x7BDF, vk10));
       vo8ACEp2 = wasm_f32x4_add(vo8ACEp2, wasm_f32x4_mul(vi2x7BDF, vk20));

       v128_t vo = wasm_f32x4_add(vo8ACEp0, vo8ACEp1);
       vo = wasm_f32x4_add(vo, vo8ACEp2);

       $if X86:
          vo = wasm_v128_bitselect(vmin, vo, wasm_f32x4_lt(vo, vmin));
          vo = wasm_v128_bitselect(vo, vmax, wasm_f32x4_le(vo, vmax));
       $else:
          vo = wasm_f32x4_max(vo, vmin);
          vo = wasm_f32x4_min(vo, vmax);

       wasm_v128_store(output, vo);
       output += 4;
     }
     // Potentially process the last block of 0..7 pixels.
     assert(w < 8 * sizeof(float));
     if XNN_LIKELY(w != 0) {
       v128_t vo8ACEp0 = vbias;

       const v128_t vi0x89AB = wasm_v128_load(i0);
       const v128_t vi1x89AB = wasm_v128_load(i1);
       const v128_t vi2x89AB = wasm_v128_load(i2);

       const v128_t vi0xCDEF = wasm_v128_load(i0 + 4);
       const v128_t vi1xCDEF = wasm_v128_load(i1 + 4);
       const v128_t vi2xCDEF = wasm_v128_load(i2 + 4);

       const v128_t vi0x8ACE = wasm_v128_and(vmask_even, concat_even_f32(vi0x89AB, vi0xCDEF));
       const v128_t vi0x9BDF = wasm_v128_and(vmask_odd,  concat_odd_f32(vi0x89AB, vi0xCDEF));
       const v128_t vi1x8ACE = wasm_v128_and(vmask_even, concat_even_f32(vi1x89AB, vi1xCDEF));
       const v128_t vi1x9BDF = wasm_v128_and(vmask_odd,  concat_odd_f32(vi1x89AB, vi1xCDEF));
       const v128_t vi2x8ACE = wasm_v128_and(vmask_even, concat_even_f32(vi2x89AB, vi2xCDEF));
       const v128_t vi2x9BDF = wasm_v128_and(vmask_odd,  concat_odd_f32(vi2x89AB, vi2xCDEF));

       vo8ACEp0 = wasm_f32x4_add(vo8ACEp0, wasm_f32x4_mul(vi0x8ACE, vk01));
       v128_t vo8ACEp1 = wasm_f32x4_mul(vi1x8ACE, vk11);
       v128_t vo8ACEp2 = wasm_f32x4_mul(vi2x8ACE, vk21);

       const v128_t vi0xF9BD = rotright_f32(vi0x9BDF);
       const v128_t vi1xF9BD = rotright_f32(vi1x9BDF);
       const v128_t vi2xF9BD = rotright_f32(vi2x9BDF);

       vo8ACEp0 = wasm_f32x4_add(vo8ACEp0, wasm_f32x4_mul(vi0x9BDF, vk02));
       vo8ACEp1 = wasm_f32x4_add(vo8ACEp1, wasm_f32x4_mul(vi1x9BDF, vk12));
       vo8ACEp2 = wasm_f32x4_add(vo8ACEp2, wasm_f32x4_mul(vi2x9BDF, vk22));

       const v128_t vi0x7BDF = movess_f32(vi0xF9BD, vi0x7531);
       const v128_t vi1x7BDF = movess_f32(vi1xF9BD, vi1x7531);
       const v128_t vi2x7BDF = movess_f32(vi2xF9BD, vi2x7531);

       vo8ACEp0 = wasm_f32x4_add(vo8ACEp0, wasm_f32x4_mul(vi0x7BDF, vk00));
       vo8ACEp1 = wasm_f32x4_add(vo8ACEp1, wasm_f32x4_mul(vi1x7BDF, vk10));
       vo8ACEp2 = wasm_f32x4_add(vo8ACEp2, wasm_f32x4_mul(vi2x7BDF, vk20));

       v128_t vo = wasm_f32x4_add(vo8ACEp0, vo8ACEp1);
       vo = wasm_f32x4_add(vo, vo8ACEp2);

       $if X86:
          vo = wasm_v128_bitselect(vmin, vo, wasm_f32x4_lt(vo, vmin));
          vo = wasm_v128_bitselect(vo, vmax, wasm_f32x4_le(vo, vmax));
       $else:
          vo = wasm_f32x4_max(vo, vmin);
          vo = wasm_f32x4_min(vo, vmax);

       if (w == 7 * sizeof(float)) {
         wasm_v128_store(output, vo);
         output += 4;
       } else {
         w += 1 * sizeof(float);
         if (w & (4 * sizeof(float))) {
           *((double*) output) = wasm_f64x2_extract_lane(vo, 0);
           output += 2;
           vo = concat_hi_f32(vo, vo);
         }
         if (w & (2 * sizeof(float))) {
           *output = wasm_f32x4_extract_lane(vo, 0);
           output += 1;
         }
       }
     }

     i0 = (const float*) ((uintptr_t) i2 - input_decrement);
     i1 = (const float*) ((uintptr_t) i0 + input_width);
     i2 = (const float*) ((uintptr_t) i1 + input_width);

     output_height -= 1;
     padded_input_height -= 2;
   } while (output_height != 0);
 }
	// 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_even_f32(v128_t a, v128_t b) {
	return wasm_v32x4_shuffle(a, b, 0, 2, 4+0, 4+2);
	}

	static v128_t concat_odd_f32(v128_t a, v128_t b) {
	return wasm_v32x4_shuffle(a, b, 1, 3, 4+1, 4+3);
	}

	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 movess_f32(v128_t a, v128_t b) {
	return wasm_v32x4_shuffle(a, b, 4, 1, 2, 3);
	}

	$ARCH_SUFFIX = "_x86" if X86 else "_arm"

	void xnn_f32_dwconv2d_chw_ukernel_3x3s2p1__wasmsimd${ARCH_SUFFIX}_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 >= 0);
	assert(padding_top <= 1);

	const v128_t vmask_even = wasm_v128_load(params->scalar.mask_even);
	const v128_t vmask_odd = wasm_v128_load(params->scalar.mask_odd);
	const v128_t vmax = wasm_v32x4_load_splat(&params->scalar.max);
	const v128_t vmin = wasm_v32x4_load_splat(&params->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_down_po2(input_width, 4 /* SIMD output width / 2 /* subsampling / sizeof(float));

	const float* i0 = (const float*) ((uintptr_t) input - ((-padding_top) & input_width));
	const float* i1 = (const float*) ((uintptr_t) i0 + input_width);
	if XNN_UNPREDICTABLE(padding_top != 0) {
	i0 = zero;
	}
	const float* i2 = (const float*) ((uintptr_t) i1 + input_width);

	size_t padded_input_height = input_height + padding_top + 1 /* padding bottom */;
	size_t output_height = (padded_input_height - 3 /* kernel size / + 2 / subsampling */) / 2;
	do {
	if XNN_UNPREDICTABLE(padded_input_height <= 3) {
	i2 = zero;
	}

	v128_t vi0x7531 = wasm_f32x4_splat(0.0f);
	v128_t vi1x7531 = wasm_f32x4_splat(0.0f);
	v128_t vi2x7531 = wasm_f32x4_splat(0.0f);

	size_t w = input_width;
	for (; w >= 8 * sizeof(float); w -= 8 * sizeof(float)) {
	v128_t vo8ACEp0 = vbias;

	const v128_t vi0x89AB = wasm_v128_load(i0);
	const v128_t vi1x89AB = wasm_v128_load(i1);
	const v128_t vi2x89AB = wasm_v128_load(i2);

	const v128_t vi0xCDEF = wasm_v128_load(i0 + 4);
	i0 += 8;
	const v128_t vi1xCDEF = wasm_v128_load(i1 + 4);
	i1 += 8;
	const v128_t vi2xCDEF = wasm_v128_load(i2 + 4);
	i2 += 8;

	const v128_t vi0x8ACE = concat_even_f32(vi0x89AB, vi0xCDEF);
	const v128_t vi0x9BDF = concat_odd_f32(vi0x89AB, vi0xCDEF);
	const v128_t vi1x8ACE = concat_even_f32(vi1x89AB, vi1xCDEF);
	const v128_t vi1x9BDF = concat_odd_f32(vi1x89AB, vi1xCDEF);
	const v128_t vi2x8ACE = concat_even_f32(vi2x89AB, vi2xCDEF);
	const v128_t vi2x9BDF = concat_odd_f32(vi2x89AB, vi2xCDEF);

	vo8ACEp0 = wasm_f32x4_add(vo8ACEp0, wasm_f32x4_mul(vi0x8ACE, vk01));
	v128_t vo8ACEp1 = wasm_f32x4_mul(vi1x8ACE, vk11);
	v128_t vo8ACEp2 = wasm_f32x4_mul(vi2x8ACE, vk21);

	const v128_t vi0xF9BD = rotright_f32(vi0x9BDF);
	const v128_t vi1xF9BD = rotright_f32(vi1x9BDF);
	const v128_t vi2xF9BD = rotright_f32(vi2x9BDF);

	vo8ACEp0 = wasm_f32x4_add(vo8ACEp0, wasm_f32x4_mul(vi0x9BDF, vk02));
	vo8ACEp1 = wasm_f32x4_add(vo8ACEp1, wasm_f32x4_mul(vi1x9BDF, vk12));
	vo8ACEp2 = wasm_f32x4_add(vo8ACEp2, wasm_f32x4_mul(vi2x9BDF, vk22));

	const v128_t vi0x7BDF = movess_f32(vi0xF9BD, vi0x7531);
	const v128_t vi1x7BDF = movess_f32(vi1xF9BD, vi1x7531);
	const v128_t vi2x7BDF = movess_f32(vi2xF9BD, vi2x7531);

	vi0x7531 = vi0xF9BD;
	vi1x7531 = vi1xF9BD;
	vi2x7531 = vi2xF9BD;

	vo8ACEp0 = wasm_f32x4_add(vo8ACEp0, wasm_f32x4_mul(vi0x7BDF, vk00));
	vo8ACEp1 = wasm_f32x4_add(vo8ACEp1, wasm_f32x4_mul(vi1x7BDF, vk10));
	vo8ACEp2 = wasm_f32x4_add(vo8ACEp2, wasm_f32x4_mul(vi2x7BDF, vk20));

	v128_t vo = wasm_f32x4_add(vo8ACEp0, vo8ACEp1);
	vo = wasm_f32x4_add(vo, vo8ACEp2);

	$if X86:
	vo = wasm_v128_bitselect(vmin, vo, wasm_f32x4_lt(vo, vmin));
	vo = wasm_v128_bitselect(vo, vmax, wasm_f32x4_le(vo, vmax));
	$else:
	vo = wasm_f32x4_max(vo, vmin);
	vo = wasm_f32x4_min(vo, vmax);

	wasm_v128_store(output, vo);
	output += 4;
	}
	// Potentially process the last block of 0..7 pixels.
	assert(w < 8 * sizeof(float));
	if XNN_LIKELY(w != 0) {
	v128_t vo8ACEp0 = vbias;

	const v128_t vi0x89AB = wasm_v128_load(i0);
	const v128_t vi1x89AB = wasm_v128_load(i1);
	const v128_t vi2x89AB = wasm_v128_load(i2);

	const v128_t vi0xCDEF = wasm_v128_load(i0 + 4);
	const v128_t vi1xCDEF = wasm_v128_load(i1 + 4);
	const v128_t vi2xCDEF = wasm_v128_load(i2 + 4);

	const v128_t vi0x8ACE = wasm_v128_and(vmask_even, concat_even_f32(vi0x89AB, vi0xCDEF));
	const v128_t vi0x9BDF = wasm_v128_and(vmask_odd, concat_odd_f32(vi0x89AB, vi0xCDEF));
	const v128_t vi1x8ACE = wasm_v128_and(vmask_even, concat_even_f32(vi1x89AB, vi1xCDEF));
	const v128_t vi1x9BDF = wasm_v128_and(vmask_odd, concat_odd_f32(vi1x89AB, vi1xCDEF));
	const v128_t vi2x8ACE = wasm_v128_and(vmask_even, concat_even_f32(vi2x89AB, vi2xCDEF));
	const v128_t vi2x9BDF = wasm_v128_and(vmask_odd, concat_odd_f32(vi2x89AB, vi2xCDEF));

	vo8ACEp0 = wasm_f32x4_add(vo8ACEp0, wasm_f32x4_mul(vi0x8ACE, vk01));
	v128_t vo8ACEp1 = wasm_f32x4_mul(vi1x8ACE, vk11);
	v128_t vo8ACEp2 = wasm_f32x4_mul(vi2x8ACE, vk21);

	const v128_t vi0xF9BD = rotright_f32(vi0x9BDF);
	const v128_t vi1xF9BD = rotright_f32(vi1x9BDF);
	const v128_t vi2xF9BD = rotright_f32(vi2x9BDF);

	vo8ACEp0 = wasm_f32x4_add(vo8ACEp0, wasm_f32x4_mul(vi0x9BDF, vk02));
	vo8ACEp1 = wasm_f32x4_add(vo8ACEp1, wasm_f32x4_mul(vi1x9BDF, vk12));
	vo8ACEp2 = wasm_f32x4_add(vo8ACEp2, wasm_f32x4_mul(vi2x9BDF, vk22));

	const v128_t vi0x7BDF = movess_f32(vi0xF9BD, vi0x7531);
	const v128_t vi1x7BDF = movess_f32(vi1xF9BD, vi1x7531);
	const v128_t vi2x7BDF = movess_f32(vi2xF9BD, vi2x7531);

	vo8ACEp0 = wasm_f32x4_add(vo8ACEp0, wasm_f32x4_mul(vi0x7BDF, vk00));
	vo8ACEp1 = wasm_f32x4_add(vo8ACEp1, wasm_f32x4_mul(vi1x7BDF, vk10));
	vo8ACEp2 = wasm_f32x4_add(vo8ACEp2, wasm_f32x4_mul(vi2x7BDF, vk20));

	v128_t vo = wasm_f32x4_add(vo8ACEp0, vo8ACEp1);
	vo = wasm_f32x4_add(vo, vo8ACEp2);

	$if X86:
	vo = wasm_v128_bitselect(vmin, vo, wasm_f32x4_lt(vo, vmin));
	vo = wasm_v128_bitselect(vo, vmax, wasm_f32x4_le(vo, vmax));
	$else:
	vo = wasm_f32x4_max(vo, vmin);
	vo = wasm_f32x4_min(vo, vmax);

	if (w == 7 * sizeof(float)) {
	wasm_v128_store(output, vo);
	output += 4;
	} else {
	w += 1 * sizeof(float);
	if (w & (4 * sizeof(float))) {
	((double) output) = wasm_f64x2_extract_lane(vo, 0);
	output += 2;
	vo = concat_hi_f32(vo, vo);
	}
	if (w & (2 * sizeof(float))) {
	*output = wasm_f32x4_extract_lane(vo, 0);
	output += 1;
	}
	}
	}

	i0 = (const float*) ((uintptr_t) i2 - input_decrement);
	i1 = (const float*) ((uintptr_t) i0 + input_width);
	i2 = (const float*) ((uintptr_t) i1 + input_width);

	output_height -= 1;
	padded_input_height -= 2;
	} while (output_height != 0);
	}