src/f32-dwconv-chw/3x3p1-psimd.c - platform/external/XNNPACK - Gitiles

 // 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 <assert.h>

 #include <psimd.h>

 #include <xnnpack/dwconv.h>
 #include <xnnpack/math.h>

 PSIMD_INTRINSIC psimd_f32 rotright_f32(psimd_f32 a) {
   #if defined(__clang__)
     return __builtin_shufflevector(a, a, 3, 0, 1, 2);
   #else
     return __builtin_shuffle(a, (psimd_s32){3, 0, 1, 2});
   #endif  // defined(__clang__)
 }

 PSIMD_INTRINSIC psimd_f32 rotleft_f32(psimd_f32 a) {
   #if defined(__clang__)
     return __builtin_shufflevector(a, a, 1, 2, 3, 0);
   #else
     return __builtin_shuffle(a, (psimd_s32){1, 2, 3, 0});
   #endif  // defined(__clang__)
 }

 PSIMD_INTRINSIC psimd_f32 movess_f32(psimd_f32 a, psimd_f32 b) {
   #if defined(__clang__)
     return __builtin_shufflevector(a, b, 4, 1, 2, 3);
   #else
     return __builtin_shuffle(a, b, (psimd_s32){4, 1, 2, 3});
   #endif  // defined(__clang__)
 }

 void xnn_f32_dwconv_chw_ukernel_3x3p1__psimd(
     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_width != 0);
   assert(input_height != 0);
   assert(padding_top == 1);

   const size_t input_tuple_stride = 4 * sizeof(float);
   const size_t output_tuple_stride = 4 * sizeof(float);
   const size_t input_width_stride = input_width * sizeof(float);
   const size_t output_width = input_width;
   const size_t output_width_stride = output_width * sizeof(float);

   const size_t padded_input_height = input_height + padding_top + 1 /* padding_bottom */;
   const size_t output_height = padded_input_height - 3 + 1;

   const psimd_s32 vmask = psimd_load_s32(params->scalar.mask);
   const psimd_f32 vmax = psimd_load_splat_f32(&params->scalar.max);
   const psimd_f32 vmin = psimd_load_splat_f32(&params->scalar.min);

   const size_t input_width_decrement = round_up_po2(input_width, 4) / 4 * input_tuple_stride;
   const size_t input_width_increment = input_width_stride - input_width_decrement;
   const size_t output_width_increment = output_width_stride - (input_width - 1) / 4 * output_tuple_stride;

   const float* i0 = zero;
   const float* i1 = input;
   const float* i2 = (const float*) ((uintptr_t) i1 + input_width_stride);

   if (input_height == 1) {
     i2 = zero;
   }

   const psimd_f32 vbias = psimd_load_splat_f32(weights);
   const psimd_f32 vk00 = psimd_load_splat_f32(weights + 1);
   const psimd_f32 vk01 = psimd_load_splat_f32(weights + 2);
   const psimd_f32 vk02 = psimd_load_splat_f32(weights + 3);
   const psimd_f32 vk10 = psimd_load_splat_f32(weights + 4);
   const psimd_f32 vk11 = psimd_load_splat_f32(weights + 5);
   const psimd_f32 vk12 = psimd_load_splat_f32(weights + 6);
   const psimd_f32 vk20 = psimd_load_splat_f32(weights + 7);
   const psimd_f32 vk21 = psimd_load_splat_f32(weights + 8);
   const psimd_f32 vk22 = psimd_load_splat_f32(weights + 9);

   size_t m = output_height;
   do {
     // vi0x3012 = ( vi02, vi01, vi00, vi03 )
     psimd_f32 vi0x3012 = psimd_zero_f32();
     // vi1x3012 = ( vi12, vi11, vi10, vi13 )
     psimd_f32 vi1x3012 = psimd_zero_f32();
     // vi2x3012 = ( vi22, vi21, vi20, vi13 )
     psimd_f32 vi2x3012 = psimd_zero_f32();
     // vi0x4567 = ( vi07, vi06, vi05, vi04 )
     psimd_f32 vi0x4567 = psimd_load_f32(i0);
     i0 = (const float*) ((uintptr_t) i0 + input_tuple_stride);
     // vi1x4567 = ( vi17, vi16, vi15, vi14 )
     psimd_f32 vi1x4567 = psimd_load_f32(i1);
     i1 = (const float*) ((uintptr_t) i1 + input_tuple_stride);
     // vi2x4567 = ( vi27, vi26, vi25, vi24 )
     psimd_f32 vi2x4567 = psimd_load_f32(i2);
     i2 = (const float*) ((uintptr_t) i2 + input_tuple_stride);

     size_t k = input_width;
     for (; k > 4; k -= 4) {
       psimd_f32 vo4567p0 = vbias;

       // vi0x89AB = ( vi0B, vi0A, vi09, vi08 )
       const psimd_f32 vi0x89AB = psimd_load_f32(i0);
       i0 = (const float*) ((uintptr_t) i0 + input_tuple_stride);
       // vi1x89AB = ( vi1B, vi0A, vi09, vi08 )
       const psimd_f32 vi1x89AB = psimd_load_f32(i1);
       i1 = (const float*) ((uintptr_t) i1 + input_tuple_stride);
       // vi2x89AB = ( vi2B, vi0A, vi09, vi08 )
       const psimd_f32 vi2x89AB = psimd_load_f32(i2);
       i2 = (const float*) ((uintptr_t) i2 + input_tuple_stride);

       // vi0x7456 = ( vi06, vi05, vi04, vi07 )
       const psimd_f32 vi0x7456 = rotright_f32(vi0x4567);
       // vi1x7456 = ( vi16, vi15, vi14, vi17 )
       const psimd_f32 vi1x7456 = rotright_f32(vi1x4567);
       // vi2x7456 = ( vi26, vi25, vi24, vi27 )
       const psimd_f32 vi2x7456 = rotright_f32(vi2x4567);

       vo4567p0 = psimd_qfma_f32(vo4567p0, vi0x4567, vk01);
       psimd_f32 vo4567p1 = psimd_mul_f32(vi1x4567, vk11);
       psimd_f32 vo4567p2 = psimd_mul_f32(vi2x4567, vk21);

       // vi0x3456 = ( vi06, vi05, vi04, vi03 )
       const psimd_f32 vi0x3456 = movess_f32(vi0x7456, vi0x3012);
       // vi1x3456 = ( vi16, vi15, vi14, vi13 )
       const psimd_f32 vi1x3456 = movess_f32(vi1x7456, vi1x3012);
       // vi2x3456 = ( vi26, vi25, vi24, vi23 )
       const psimd_f32 vi2x3456 = movess_f32(vi2x7456, vi2x3012);

       vo4567p0 = psimd_qfma_f32(vo4567p0, vi0x3456, vk00);
       vo4567p1 = psimd_qfma_f32(vo4567p1, vi1x3456, vk10);
       vo4567p2 = psimd_qfma_f32(vo4567p2, vi2x3456, vk20);

       vi0x3012 = vi0x7456;
       vi1x3012 = vi1x7456;
       vi2x3012 = vi2x7456;

       // vi0x8567 = ( vi07, vi06, vi05, vi08 )
       const psimd_f32 vi0x8567 = movess_f32(vi0x4567, vi0x89AB);
       // vi1x8567 = ( vi17, vi16, vi15, vi18 )
       const psimd_f32 vi1x8567 = movess_f32(vi1x4567, vi1x89AB);
       // vi2x8567 = ( vi27, vi26, vi25, vi28 )
       const psimd_f32 vi2x8567 = movess_f32(vi2x4567, vi2x89AB);

       // vi0x5678 = ( vi08, vi07, vi06, vi05 )
       const psimd_f32 vi0x5678 = rotleft_f32(vi0x8567);
       // vi1x5678 = ( vi18, vi17, vi16, vi15 )
       const psimd_f32 vi1x5678 = rotleft_f32(vi1x8567);
       // vi2x5678 = ( vi28, vi27, vi26, vi25 )
       const psimd_f32 vi2x5678 = rotleft_f32(vi2x8567);

       vo4567p0 = psimd_qfma_f32(vo4567p0, vi0x5678, vk02);
       vo4567p1 = psimd_qfma_f32(vo4567p1, vi1x5678, vk12);
       vo4567p2 = psimd_qfma_f32(vo4567p2, vi2x5678, vk22);

       vi0x4567 = vi0x89AB;
       vi1x4567 = vi1x89AB;
       vi2x4567 = vi2x89AB;

       psimd_f32 vo = psimd_add_f32(vo4567p0, vo4567p1);
       vo = psimd_add_f32(vo, vo4567p2);

       vo = psimd_max_f32(vo, vmin);
       vo = psimd_min_f32(vo, vmax);

       psimd_store_f32(output, vo);
       output = (float*) ((uintptr_t) output + output_tuple_stride);
     }
     // Always process the last block of 1..4 pixels.
     assert(k >= 1);
     assert(k <= 4);
     {
       psimd_f32 vo4567p0 = vbias;

       vi0x4567 = psimd_andmask_f32(vmask, vi0x4567);
       vi1x4567 = psimd_andmask_f32(vmask, vi1x4567);
       vi2x4567 = psimd_andmask_f32(vmask, vi2x4567);

       // vi0x7456 = ( vi06, vi05, vi04, vi07 )
       const psimd_f32 vi0x7456 = rotright_f32(vi0x4567);
       // vi1x7456 = ( vi16, vi15, vi14, vi17 )
       const psimd_f32 vi1x7456 = rotright_f32(vi1x4567);
       // vi2x7456 = ( vi26, vi25, vi24, vi27 )
       const psimd_f32 vi2x7456 = rotright_f32(vi2x4567);

       vo4567p0 = psimd_qfma_f32(vo4567p0, vi0x4567, vk01);
       psimd_f32 vo4567p1 = psimd_mul_f32(vi1x4567, vk11);
       psimd_f32 vo4567p2 = psimd_mul_f32(vi2x4567, vk21);

       // vi0x3456 = ( vi06, vi05, vi04, vi03 )
       const psimd_f32 vi0x3456 = movess_f32(vi0x7456, vi0x3012);
       // vi1x3456 = ( vi16, vi15, vi14, vi13 )
       const psimd_f32 vi1x3456 = movess_f32(vi1x7456, vi1x3012);
       // vi2x3456 = ( vi26, vi25, vi24, vi23 )
       const psimd_f32 vi2x3456 = movess_f32(vi2x7456, vi2x3012);

       vo4567p0 = psimd_qfma_f32(vo4567p0, vi0x3456, vk00);
       vo4567p1 = psimd_qfma_f32(vo4567p1, vi1x3456, vk10);
       vo4567p2 = psimd_qfma_f32(vo4567p2, vi2x3456, vk20);

       const psimd_f32 vzero = psimd_zero_f32();
       // vi0x8567 = ( vi07, vi06, vi05, 0.0 )
       const psimd_f32 vi0x8567 = movess_f32(vi0x4567, vzero);
       // vi1x8567 = ( vi17, vi16, vi15, 0.0 )
       const psimd_f32 vi1x8567 = movess_f32(vi1x4567, vzero);
       // vi2x8567 = ( vi27, vi26, vi25, 0.0 )
       const psimd_f32 vi2x8567 = movess_f32(vi2x4567, vzero);

       // vi0x5678 = ( vi08, vi07, vi06, vi05 )
       const psimd_f32 vi0x5678 = rotleft_f32(vi0x8567);
       // vi1x5678 = ( vi18, vi17, vi16, vi15 )
       const psimd_f32 vi1x5678 = rotleft_f32(vi1x8567);
       // vi2x5678 = ( vi28, vi27, vi26, vi25 )
       const psimd_f32 vi2x5678 = rotleft_f32(vi2x8567);

       vo4567p0 = psimd_qfma_f32(vo4567p0, vi0x5678, vk02);
       vo4567p1 = psimd_qfma_f32(vo4567p1, vi1x5678, vk12);
       vo4567p2 = psimd_qfma_f32(vo4567p2, vi2x5678, vk22);

       psimd_f32 vo = psimd_add_f32(vo4567p0, vo4567p1);
       vo = psimd_add_f32(vo, vo4567p2);

       vo = psimd_max_f32(vo, vmin);
       vo = psimd_min_f32(vo, vmax);

       if XNN_LIKELY(k == 4) {
         psimd_store_f32(output, vo);
       } else {
         float* output_lo = output;
         if (k & 2) {
           psimd_store2_f32(output_lo, vo);
           output_lo += 2;
           vo = psimd_concat_hi_f32(vo, vo);
         }
         if (k & 1) {
           psimd_store1_f32(output_lo, vo);
         }
       }
     }

     i0 = (const float*) ((uintptr_t) i1 - input_width_decrement);
     i1 = (const float*) ((uintptr_t) i1 + input_width_increment);
     i2 = (const float*) ((uintptr_t) i2 + input_width_increment);
     output = (float*) ((uintptr_t) output + output_width_increment);
     m -= 1;
     if (m == 1) {
       i2 = zero;
     }
   } while (m != 0);
 }
	// 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 <assert.h>

	#include <psimd.h>

	#include <xnnpack/dwconv.h>
	#include <xnnpack/math.h>

	PSIMD_INTRINSIC psimd_f32 rotright_f32(psimd_f32 a) {
	#if defined(__clang__)
	return __builtin_shufflevector(a, a, 3, 0, 1, 2);
	#else
	return __builtin_shuffle(a, (psimd_s32){3, 0, 1, 2});
	#endif // defined(__clang__)
	}

	PSIMD_INTRINSIC psimd_f32 rotleft_f32(psimd_f32 a) {
	#if defined(__clang__)
	return __builtin_shufflevector(a, a, 1, 2, 3, 0);
	#else
	return __builtin_shuffle(a, (psimd_s32){1, 2, 3, 0});
	#endif // defined(__clang__)
	}

	PSIMD_INTRINSIC psimd_f32 movess_f32(psimd_f32 a, psimd_f32 b) {
	#if defined(__clang__)
	return __builtin_shufflevector(a, b, 4, 1, 2, 3);
	#else
	return __builtin_shuffle(a, b, (psimd_s32){4, 1, 2, 3});
	#endif // defined(__clang__)
	}

	void xnn_f32_dwconv_chw_ukernel_3x3p1__psimd(
	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_width != 0);
	assert(input_height != 0);
	assert(padding_top == 1);

	const size_t input_tuple_stride = 4 * sizeof(float);
	const size_t output_tuple_stride = 4 * sizeof(float);
	const size_t input_width_stride = input_width * sizeof(float);
	const size_t output_width = input_width;
	const size_t output_width_stride = output_width * sizeof(float);

	const size_t padded_input_height = input_height + padding_top + 1 /* padding_bottom */;
	const size_t output_height = padded_input_height - 3 + 1;

	const psimd_s32 vmask = psimd_load_s32(params->scalar.mask);
	const psimd_f32 vmax = psimd_load_splat_f32(&params->scalar.max);
	const psimd_f32 vmin = psimd_load_splat_f32(&params->scalar.min);

	const size_t input_width_decrement = round_up_po2(input_width, 4) / 4 * input_tuple_stride;
	const size_t input_width_increment = input_width_stride - input_width_decrement;
	const size_t output_width_increment = output_width_stride - (input_width - 1) / 4 * output_tuple_stride;

	const float* i0 = zero;
	const float* i1 = input;
	const float* i2 = (const float*) ((uintptr_t) i1 + input_width_stride);

	if (input_height == 1) {
	i2 = zero;
	}

	const psimd_f32 vbias = psimd_load_splat_f32(weights);
	const psimd_f32 vk00 = psimd_load_splat_f32(weights + 1);
	const psimd_f32 vk01 = psimd_load_splat_f32(weights + 2);
	const psimd_f32 vk02 = psimd_load_splat_f32(weights + 3);
	const psimd_f32 vk10 = psimd_load_splat_f32(weights + 4);
	const psimd_f32 vk11 = psimd_load_splat_f32(weights + 5);
	const psimd_f32 vk12 = psimd_load_splat_f32(weights + 6);
	const psimd_f32 vk20 = psimd_load_splat_f32(weights + 7);
	const psimd_f32 vk21 = psimd_load_splat_f32(weights + 8);
	const psimd_f32 vk22 = psimd_load_splat_f32(weights + 9);

	size_t m = output_height;
	do {
	// vi0x3012 = ( vi02, vi01, vi00, vi03 )
	psimd_f32 vi0x3012 = psimd_zero_f32();
	// vi1x3012 = ( vi12, vi11, vi10, vi13 )
	psimd_f32 vi1x3012 = psimd_zero_f32();
	// vi2x3012 = ( vi22, vi21, vi20, vi13 )
	psimd_f32 vi2x3012 = psimd_zero_f32();
	// vi0x4567 = ( vi07, vi06, vi05, vi04 )
	psimd_f32 vi0x4567 = psimd_load_f32(i0);
	i0 = (const float*) ((uintptr_t) i0 + input_tuple_stride);
	// vi1x4567 = ( vi17, vi16, vi15, vi14 )
	psimd_f32 vi1x4567 = psimd_load_f32(i1);
	i1 = (const float*) ((uintptr_t) i1 + input_tuple_stride);
	// vi2x4567 = ( vi27, vi26, vi25, vi24 )
	psimd_f32 vi2x4567 = psimd_load_f32(i2);
	i2 = (const float*) ((uintptr_t) i2 + input_tuple_stride);

	size_t k = input_width;
	for (; k > 4; k -= 4) {
	psimd_f32 vo4567p0 = vbias;

	// vi0x89AB = ( vi0B, vi0A, vi09, vi08 )
	const psimd_f32 vi0x89AB = psimd_load_f32(i0);
	i0 = (const float*) ((uintptr_t) i0 + input_tuple_stride);
	// vi1x89AB = ( vi1B, vi0A, vi09, vi08 )
	const psimd_f32 vi1x89AB = psimd_load_f32(i1);
	i1 = (const float*) ((uintptr_t) i1 + input_tuple_stride);
	// vi2x89AB = ( vi2B, vi0A, vi09, vi08 )
	const psimd_f32 vi2x89AB = psimd_load_f32(i2);
	i2 = (const float*) ((uintptr_t) i2 + input_tuple_stride);

	// vi0x7456 = ( vi06, vi05, vi04, vi07 )
	const psimd_f32 vi0x7456 = rotright_f32(vi0x4567);
	// vi1x7456 = ( vi16, vi15, vi14, vi17 )
	const psimd_f32 vi1x7456 = rotright_f32(vi1x4567);
	// vi2x7456 = ( vi26, vi25, vi24, vi27 )
	const psimd_f32 vi2x7456 = rotright_f32(vi2x4567);

	vo4567p0 = psimd_qfma_f32(vo4567p0, vi0x4567, vk01);
	psimd_f32 vo4567p1 = psimd_mul_f32(vi1x4567, vk11);
	psimd_f32 vo4567p2 = psimd_mul_f32(vi2x4567, vk21);

	// vi0x3456 = ( vi06, vi05, vi04, vi03 )
	const psimd_f32 vi0x3456 = movess_f32(vi0x7456, vi0x3012);
	// vi1x3456 = ( vi16, vi15, vi14, vi13 )
	const psimd_f32 vi1x3456 = movess_f32(vi1x7456, vi1x3012);
	// vi2x3456 = ( vi26, vi25, vi24, vi23 )
	const psimd_f32 vi2x3456 = movess_f32(vi2x7456, vi2x3012);

	vo4567p0 = psimd_qfma_f32(vo4567p0, vi0x3456, vk00);
	vo4567p1 = psimd_qfma_f32(vo4567p1, vi1x3456, vk10);
	vo4567p2 = psimd_qfma_f32(vo4567p2, vi2x3456, vk20);

	vi0x3012 = vi0x7456;
	vi1x3012 = vi1x7456;
	vi2x3012 = vi2x7456;

	// vi0x8567 = ( vi07, vi06, vi05, vi08 )
	const psimd_f32 vi0x8567 = movess_f32(vi0x4567, vi0x89AB);
	// vi1x8567 = ( vi17, vi16, vi15, vi18 )
	const psimd_f32 vi1x8567 = movess_f32(vi1x4567, vi1x89AB);
	// vi2x8567 = ( vi27, vi26, vi25, vi28 )
	const psimd_f32 vi2x8567 = movess_f32(vi2x4567, vi2x89AB);

	// vi0x5678 = ( vi08, vi07, vi06, vi05 )
	const psimd_f32 vi0x5678 = rotleft_f32(vi0x8567);
	// vi1x5678 = ( vi18, vi17, vi16, vi15 )
	const psimd_f32 vi1x5678 = rotleft_f32(vi1x8567);
	// vi2x5678 = ( vi28, vi27, vi26, vi25 )
	const psimd_f32 vi2x5678 = rotleft_f32(vi2x8567);

	vo4567p0 = psimd_qfma_f32(vo4567p0, vi0x5678, vk02);
	vo4567p1 = psimd_qfma_f32(vo4567p1, vi1x5678, vk12);
	vo4567p2 = psimd_qfma_f32(vo4567p2, vi2x5678, vk22);

	vi0x4567 = vi0x89AB;
	vi1x4567 = vi1x89AB;
	vi2x4567 = vi2x89AB;

	psimd_f32 vo = psimd_add_f32(vo4567p0, vo4567p1);
	vo = psimd_add_f32(vo, vo4567p2);

	vo = psimd_max_f32(vo, vmin);
	vo = psimd_min_f32(vo, vmax);

	psimd_store_f32(output, vo);
	output = (float*) ((uintptr_t) output + output_tuple_stride);
	}
	// Always process the last block of 1..4 pixels.
	assert(k >= 1);
	assert(k <= 4);
	{
	psimd_f32 vo4567p0 = vbias;

	vi0x4567 = psimd_andmask_f32(vmask, vi0x4567);
	vi1x4567 = psimd_andmask_f32(vmask, vi1x4567);
	vi2x4567 = psimd_andmask_f32(vmask, vi2x4567);

	// vi0x7456 = ( vi06, vi05, vi04, vi07 )
	const psimd_f32 vi0x7456 = rotright_f32(vi0x4567);
	// vi1x7456 = ( vi16, vi15, vi14, vi17 )
	const psimd_f32 vi1x7456 = rotright_f32(vi1x4567);
	// vi2x7456 = ( vi26, vi25, vi24, vi27 )
	const psimd_f32 vi2x7456 = rotright_f32(vi2x4567);

	vo4567p0 = psimd_qfma_f32(vo4567p0, vi0x4567, vk01);
	psimd_f32 vo4567p1 = psimd_mul_f32(vi1x4567, vk11);
	psimd_f32 vo4567p2 = psimd_mul_f32(vi2x4567, vk21);

	// vi0x3456 = ( vi06, vi05, vi04, vi03 )
	const psimd_f32 vi0x3456 = movess_f32(vi0x7456, vi0x3012);
	// vi1x3456 = ( vi16, vi15, vi14, vi13 )
	const psimd_f32 vi1x3456 = movess_f32(vi1x7456, vi1x3012);
	// vi2x3456 = ( vi26, vi25, vi24, vi23 )
	const psimd_f32 vi2x3456 = movess_f32(vi2x7456, vi2x3012);

	vo4567p0 = psimd_qfma_f32(vo4567p0, vi0x3456, vk00);
	vo4567p1 = psimd_qfma_f32(vo4567p1, vi1x3456, vk10);
	vo4567p2 = psimd_qfma_f32(vo4567p2, vi2x3456, vk20);

	const psimd_f32 vzero = psimd_zero_f32();
	// vi0x8567 = ( vi07, vi06, vi05, 0.0 )
	const psimd_f32 vi0x8567 = movess_f32(vi0x4567, vzero);
	// vi1x8567 = ( vi17, vi16, vi15, 0.0 )
	const psimd_f32 vi1x8567 = movess_f32(vi1x4567, vzero);
	// vi2x8567 = ( vi27, vi26, vi25, 0.0 )
	const psimd_f32 vi2x8567 = movess_f32(vi2x4567, vzero);

	// vi0x5678 = ( vi08, vi07, vi06, vi05 )
	const psimd_f32 vi0x5678 = rotleft_f32(vi0x8567);
	// vi1x5678 = ( vi18, vi17, vi16, vi15 )
	const psimd_f32 vi1x5678 = rotleft_f32(vi1x8567);
	// vi2x5678 = ( vi28, vi27, vi26, vi25 )
	const psimd_f32 vi2x5678 = rotleft_f32(vi2x8567);

	vo4567p0 = psimd_qfma_f32(vo4567p0, vi0x5678, vk02);
	vo4567p1 = psimd_qfma_f32(vo4567p1, vi1x5678, vk12);
	vo4567p2 = psimd_qfma_f32(vo4567p2, vi2x5678, vk22);

	psimd_f32 vo = psimd_add_f32(vo4567p0, vo4567p1);
	vo = psimd_add_f32(vo, vo4567p2);

	vo = psimd_max_f32(vo, vmin);
	vo = psimd_min_f32(vo, vmax);

	if XNN_LIKELY(k == 4) {
	psimd_store_f32(output, vo);
	} else {
	float* output_lo = output;
	if (k & 2) {
	psimd_store2_f32(output_lo, vo);
	output_lo += 2;
	vo = psimd_concat_hi_f32(vo, vo);
	}
	if (k & 1) {
	psimd_store1_f32(output_lo, vo);
	}
	}
	}

	i0 = (const float*) ((uintptr_t) i1 - input_width_decrement);
	i1 = (const float*) ((uintptr_t) i1 + input_width_increment);
	i2 = (const float*) ((uintptr_t) i2 + input_width_increment);
	output = (float*) ((uintptr_t) output + output_width_increment);
	m -= 1;
	if (m == 1) {
	i2 = zero;
	}
	} while (m != 0);
	}