/**
 * vec - a tiny SIMD vector library in C99
 * 
 * Copyright (c) 2024 Paper
 * 
 * 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 "vec/impl/x86/avx2.h"
#include "vec/impl/generic.h"

#include <immintrin.h>

// this does NOT handle sign bits properly, use with caution
#define VEC_AVX2_OPERATION_8x32_16x16(op, sign) \
	do { \
		union v##sign##int8x32_impl_data *vec1d = (union v##sign##int8x32_impl_data *)&vec1; \
		union v##sign##int8x32_impl_data *vec2d = (union v##sign##int8x32_impl_data *)&vec2; \
	\
		/* unpack and multiply */ \
		__m256i dst_even = _mm256_##op##_epi16(vec1d->avx2, vec2d->avx2); \
		__m256i dst_odd = _mm256_##op##_epi16(_mm256_srli_epi16(vec1d->avx2, 8), _mm256_srli_epi16(vec2d->avx2, 8)); \
	\
		/* repack */ \
		vec1d->avx2 = _mm256_or_si256( \
			_mm256_slli_epi16(dst_odd, 8), \
			_mm256_srli_epi16(_mm256_slli_epi16(dst_even, 8), 8) \
		); \
		return vec1d->vec; \
	} while (0)

#define VEC_AVX2_OPERATION_16x16(op, sign) \
	do { \
		union v##sign##int16x16_impl_data *vec1d = (union v##sign##int16x16_impl_data *)&vec1; \
		union v##sign##int16x16_impl_data *vec2d = (union v##sign##int16x16_impl_data *)&vec2; \
	\
		/* unpack and multiply */ \
		__m256i dst_even = _mm256_##op##_epi32(vec1d->avx2, vec2d->avx2); \
		__m256i dst_odd = _mm256_##op##_epi32(_mm256_srli_epi32(vec1d->avx2, 16), _mm256_srli_epi32(vec2d->avx2, 16)); \
	\
		/* repack */ \
		vec1d->avx2 = _mm256_or_si256( \
			_mm256_slli_epi32(dst_odd, 16), \
			_mm256_srli_epi32(_mm256_slli_epi16(dst_even, 16), 16) \
		); \
		return vec1d->vec; \
	} while (0)

// multiplication

#define VEC_AVX2_MUL_8x32(sign) \
	VEC_AVX2_OPERATION_8x32_16x16(mullo, sign)

#define VEC_AVX2_MUL_16x16(sign) \
	do { \
		union v##sign##int16x16_impl_data *vec1d = (union v##sign##int16x16_impl_data *)&vec1; \
		union v##sign##int16x16_impl_data *vec2d = (union v##sign##int16x16_impl_data *)&vec2; \
	\
		vec1d->avx2 = _mm256_mullo_epi16(vec1d->avx2, vec2d->avx2); \
		return vec1d->vec; \
	} while (0)

#define VEC_AVX2_MUL_32x8(sign) \
	do { \
		union v##sign##int32x8_impl_data *vec1d = (union v##sign##int32x8_impl_data *)&vec1; \
		union v##sign##int32x8_impl_data *vec2d = (union v##sign##int32x8_impl_data *)&vec2; \
	\
		vec1d->avx2 = _mm256_mullo_epi32(vec1d->avx2, vec2d->avx2); \
		return vec1d->vec; \
	} while (0)

#define VEC_AVX2_MUL_64x4(sign) \
	do { \
		union v##sign##int64x4_impl_data *vec1d = (union v##sign##int64x4_impl_data *)&vec1; \
		union v##sign##int64x4_impl_data *vec2d = (union v##sign##int64x4_impl_data *)&vec2; \
	\
		__m256i ac = _mm256_mul_epu32(vec1d->avx2, vec2d->avx2); \
		__m256i b  = _mm256_srli_epi64(vec1d->avx2, 32); \
		__m256i bc = _mm256_mul_epu32(b, vec2d->avx2); \
		__m256i d  = _mm256_srli_epi64(vec2d->avx2, 32); \
		__m256i ad = _mm256_mul_epu32(vec1d->avx2, d); \
		__m256i hi = _mm256_add_epi64(bc, ad); \
		hi = _mm256_slli_epi64(hi, 32); \
	\
		vec1d->avx2 = _mm256_add_epi64(hi, ac); \
		return vec1d->vec; \
	} while (0)

// operations

#define VEC_AVX2_DEFINE_OPERATIONS_SIGN(sign, bits, size) \
	union v##sign##int##bits##x##size##_impl_data { \
		v##sign##int##bits##x##size vec; \
		__m256i avx2; \
	}; \
	\
	VEC_STATIC_ASSERT(VEC_ALIGNOF(__m256i) <= VEC_ALIGNOF(v##sign##int##bits##x##size), "vec: v" #sign "int" #bits "x" #size " alignment needs to be expanded to fit intrinsic type size"); \
	VEC_STATIC_ASSERT(sizeof(__m256i) <= sizeof(v##sign##int##bits##x##size), "vec: v" #sign "int" #bits "x" #size " needs to be expanded to fit intrinsic type size"); \
	\
	static v##sign##int##bits##x##size v##sign##int##bits##x##size##_avx2_load_aligned(const vec_##sign##int##bits in[size]) \
	{ \
		union v##sign##int##bits##x##size##_impl_data vec; \
		vec.avx2 = _mm256_load_si256((const __m256i *)in); \
		return vec.vec; \
	} \
	\
	static v##sign##int##bits##x##size v##sign##int##bits##x##size##_avx2_load(const vec_##sign##int##bits in[size]) \
	{ \
		union v##sign##int##bits##x##size##_impl_data vec; \
		vec.avx2 = _mm256_loadu_si256((const __m256i *)in); \
		return vec.vec; \
	} \
	\
	static void v##sign##int##bits##x##size##_avx2_store_aligned(v##sign##int##bits##x##size vec, vec_##sign##int##bits out[size]) \
	{ \
		_mm256_store_si256((__m256i *)out, ((union v##sign##int##bits##x##size##_impl_data*)&vec)->avx2); \
	} \
	\
	static void v##sign##int##bits##x##size##_avx2_store(v##sign##int##bits##x##size vec, vec_##sign##int##bits out[size]) \
	{ \
		_mm256_storeu_si256((__m256i *)out, ((union v##sign##int##bits##x##size##_impl_data*)&vec)->avx2); \
	} \
	\
	static v##sign##int##bits##x##size v##sign##int##bits##x##size##_avx2_add(v##sign##int##bits##x##size vec1, v##sign##int##bits##x##size vec2) \
	{ \
		union v##sign##int##bits##x##size##_impl_data *vec1d = (union v##sign##int##bits##x##size##_impl_data *)&vec1; \
		union v##sign##int##bits##x##size##_impl_data *vec2d = (union v##sign##int##bits##x##size##_impl_data *)&vec2; \
	\
		vec1d->avx2 = _mm256_add_epi##bits(vec1d->avx2, vec2d->avx2); \
		return vec1d->vec; \
	} \
	\
	static v##sign##int##bits##x##size v##sign##int##bits##x##size##_avx2_sub(v##sign##int##bits##x##size vec1, v##sign##int##bits##x##size vec2) \
	{ \
		union v##sign##int##bits##x##size##_impl_data *vec1d = (union v##sign##int##bits##x##size##_impl_data *)&vec1; \
		union v##sign##int##bits##x##size##_impl_data *vec2d = (union v##sign##int##bits##x##size##_impl_data *)&vec2; \
	\
		vec1d->avx2 = _mm256_sub_epi##bits(vec1d->avx2, vec2d->avx2); \
		return vec1d->vec; \
	} \
	\
	static v##sign##int##bits##x##size v##sign##int##bits##x##size##_avx2_mul(v##sign##int##bits##x##size vec1, v##sign##int##bits##x##size vec2) \
	{ \
		VEC_AVX2_MUL_##bits##x##size(sign); \
	} \
	\
	static v##sign##int##bits##x##size v##sign##int##bits##x##size##_avx2_and(v##sign##int##bits##x##size vec1, v##sign##int##bits##x##size vec2) \
	{ \
		union v##sign##int##bits##x##size##_impl_data *vec1d = (union v##sign##int##bits##x##size##_impl_data *)&vec1; \
		union v##sign##int##bits##x##size##_impl_data *vec2d = (union v##sign##int##bits##x##size##_impl_data *)&vec2; \
	\
		vec1d->avx2 = _mm256_and_si256(vec1d->avx2, vec2d->avx2); \
		return vec1d->vec; \
	} \
	\
	static v##sign##int##bits##x##size v##sign##int##bits##x##size##_avx2_or(v##sign##int##bits##x##size vec1, v##sign##int##bits##x##size vec2) \
	{ \
		union v##sign##int##bits##x##size##_impl_data *vec1d = (union v##sign##int##bits##x##size##_impl_data *)&vec1; \
		union v##sign##int##bits##x##size##_impl_data *vec2d = (union v##sign##int##bits##x##size##_impl_data *)&vec2; \
	\
		vec1d->avx2 = _mm256_or_si256(vec1d->avx2, vec2d->avx2); \
		return vec1d->vec; \
	} \
	\
	static v##sign##int##bits##x##size v##sign##int##bits##x##size##_avx2_xor(v##sign##int##bits##x##size vec1, v##sign##int##bits##x##size vec2) \
	{ \
		union v##sign##int##bits##x##size##_impl_data *vec1d = (union v##sign##int##bits##x##size##_impl_data *)&vec1; \
		union v##sign##int##bits##x##size##_impl_data *vec2d = (union v##sign##int##bits##x##size##_impl_data *)&vec2; \
	\
		vec1d->avx2 = _mm256_xor_si256(vec1d->avx2, vec2d->avx2); \
		return vec1d->vec; \
	} \
	\
	const v##sign##int##bits##x##size##_impl v##sign##int##bits##x##size##_impl_avx2 = { \
		v##sign##int##bits##x##size##_generic_splat, \
		v##sign##int##bits##x##size##_avx2_load_aligned, \
		v##sign##int##bits##x##size##_avx2_load, \
		v##sign##int##bits##x##size##_avx2_store_aligned, \
		v##sign##int##bits##x##size##_avx2_store, \
		v##sign##int##bits##x##size##_avx2_add, \
		v##sign##int##bits##x##size##_avx2_sub, \
		v##sign##int##bits##x##size##_avx2_mul, \
		v##sign##int##bits##x##size##_generic_div, \
		v##sign##int##bits##x##size##_generic_avg, \
		v##sign##int##bits##x##size##_avx2_and, \
		v##sign##int##bits##x##size##_avx2_or, \
		v##sign##int##bits##x##size##_avx2_xor, \
		v##sign##int##bits##x##size##_generic_not, \
		v##sign##int##bits##x##size##_generic_lshift, \
		v##sign##int##bits##x##size##_generic_rshift, \
		v##sign##int##bits##x##size##_generic_lrshift, \
		v##sign##int##bits##x##size##_generic_cmplt, \
		v##sign##int##bits##x##size##_generic_cmple, \
		v##sign##int##bits##x##size##_generic_cmpeq, \
		v##sign##int##bits##x##size##_generic_cmpge, \
		v##sign##int##bits##x##size##_generic_cmpgt, \
	};

#define VEC_AVX2_DEFINE_OPERATIONS(bits, size) \
	VEC_AVX2_DEFINE_OPERATIONS_SIGN( , bits, size) \
	VEC_AVX2_DEFINE_OPERATIONS_SIGN(u, bits, size)

VEC_AVX2_DEFINE_OPERATIONS(8, 32)
VEC_AVX2_DEFINE_OPERATIONS(16, 16)
VEC_AVX2_DEFINE_OPERATIONS(32, 8)
VEC_AVX2_DEFINE_OPERATIONS(64, 4)