/*
 * sha256_ni_rec.cpp
 *
 *  Created on: Jan 29, 2025
 *      Author: mad, voidxno
 *
 *  NOTE: Non-official architecture edition for Intel 15th-gen E-core (for TimeLord, VDF-verify _2x still identical).
 */

// prerequisite: length is 32 bytes (recursive sha256)
// prerequisite: length is 64 bytes, 2x 32bytes (_x2)
// optimization: https://github.com/voidxno/fast-recursive-sha256

#include <sha256_ni.h>

#include <cstring>
#include <stdexcept>

#if defined(__SHA__) || defined(_WIN32)

#include <immintrin.h>

#ifdef _WIN32
#include <intrin.h>
#endif

void recursive_sha256_ni(uint8_t* hash, const uint64_t num_iters)
{
	if(num_iters <= 0) {
		return;
	}

	alignas(64)
	static const uint32_t K64[64] = {
		0x428A2F98, 0x71374491, 0xB5C0FBCF, 0xE9B5DBA5,
		0x3956C25B, 0x59F111F1, 0x923F82A4, 0xAB1C5ED5,
		0xD807AA98, 0x12835B01, 0x243185BE, 0x550C7DC3,
		0x72BE5D74, 0x80DEB1FE, 0x9BDC06A7, 0xC19BF174,
		0xE49B69C1, 0xEFBE4786, 0x0FC19DC6, 0x240CA1CC,
		0x2DE92C6F, 0x4A7484AA, 0x5CB0A9DC, 0x76F988DA,
		0x983E5152, 0xA831C66D, 0xB00327C8, 0xBF597FC7,
		0xC6E00BF3, 0xD5A79147, 0x06CA6351, 0x14292967,
		0x27B70A85, 0x2E1B2138, 0x4D2C6DFC, 0x53380D13,
		0x650A7354, 0x766A0ABB, 0x81C2C92E, 0x92722C85,
		0xA2BFE8A1, 0xA81A664B, 0xC24B8B70, 0xC76C51A3,
		0xD192E819, 0xD6990624, 0xF40E3585, 0x106AA070,
		0x19A4C116, 0x1E376C08, 0x2748774C, 0x34B0BCB5,
		0x391C0CB3, 0x4ED8AA4A, 0x5B9CCA4F, 0x682E6FF3,
		0x748F82EE, 0x78A5636F, 0x84C87814, 0x8CC70208,
		0x90BEFFFA, 0xA4506CEB, 0xBEF9A3F7, 0xC67178F2
	};

	// shuffle mask
	const __m128i SHUF_MASK = _mm_set_epi64x(0x0c0d0e0f08090a0b,0x0405060700010203);

	// init values
	const __m128i ABEF_INIT = _mm_set_epi64x(0x6a09e667bb67ae85,0x510e527f9b05688c);
	const __m128i CDGH_INIT = _mm_set_epi64x(0x3c6ef372a54ff53a,0x1f83d9ab5be0cd19);

	// pre-calc/cache padding
	const __m128i HPAD0_CACHE = _mm_set_epi64x(0x0000000000000000,0x0000000080000000);
	const __m128i HPAD1_CACHE = _mm_set_epi64x(0x0000010000000000,0x0000000000000000);

	//-- pre-calc/static values
	const __m128i MORE0_CACHE = _mm_set_epi64x(0x5807AA985807AA98,0x550C7DC3243185BE);
	const __m128i MORE1_CACHE = _mm_set_epi64x(0x0000000000000000,0x0000000000000000);

	__m128i STATE0;
	__m128i STATE1;
	__m128i MSG;
	__m128i MSGTMP0;
	__m128i MSGTMP1;
	__m128i MSGTMP2;
	__m128i MSGTMP3;

	// init/shuffle hash
	__m128i HASH0_SAVE = _mm_loadu_si128(reinterpret_cast<__m128i*>(&hash[0]));
	__m128i HASH1_SAVE = _mm_loadu_si128(reinterpret_cast<__m128i*>(&hash[16]));
	HASH0_SAVE = _mm_shuffle_epi8(HASH0_SAVE,SHUF_MASK);
	HASH1_SAVE = _mm_shuffle_epi8(HASH1_SAVE,SHUF_MASK);

	for(uint64_t i = 0; i < num_iters; ++i) {

		// init state
		STATE0 = ABEF_INIT;
		STATE1 = CDGH_INIT;

		// rounds 0-3
		MSG = HASH0_SAVE;
		MSGTMP0 = MSG;
		MSG = _mm_add_epi32(MSG,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[0])));
		STATE1 = _mm_sha256rnds2_epu32(STATE1,STATE0,MSG);
		MSG = _mm_shuffle_epi32(MSG,0x0E);
		STATE0 = _mm_sha256rnds2_epu32(STATE0,STATE1,MSG);

		// rounds 4-7
		MSG = HASH1_SAVE;
		MSGTMP1 = MSG;
		MSG = _mm_add_epi32(MSG,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[4])));
		STATE1 = _mm_sha256rnds2_epu32(STATE1,STATE0,MSG);
		MSG = _mm_shuffle_epi32(MSG,0x0E);
		STATE0 = _mm_sha256rnds2_epu32(STATE0,STATE1,MSG);
		MSGTMP0 = _mm_sha256msg1_epu32(MSGTMP0,MSGTMP1);

		// rounds 8-11
		MSG = HPAD0_CACHE;
		MSGTMP2 = MSG;
		MSG = _mm_add_epi32(MSG,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[8])));
		STATE1 = _mm_sha256rnds2_epu32(STATE1,STATE0,MSG);
		MSG = _mm_shuffle_epi32(MSG,0x0E);
		STATE0 = _mm_sha256rnds2_epu32(STATE0,STATE1,MORE0_CACHE);
		MSGTMP1 = _mm_sha256msg1_epu32(MSGTMP1,MSGTMP2);

		// rounds 12-15
		MSG = HPAD1_CACHE;
		MSGTMP3 = MSG;
		MSG = _mm_add_epi32(MSG,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[12])));
		STATE1 = _mm_sha256rnds2_epu32(STATE1,STATE0,MSG);
		MSGTMP0 = _mm_add_epi32(MSGTMP0,MORE1_CACHE);
		MSGTMP0 = _mm_sha256msg2_epu32(MSGTMP0,MSGTMP3);
		MSG = _mm_shuffle_epi32(MSG,0x0E);
		STATE0 = _mm_sha256rnds2_epu32(STATE0,STATE1,MSG);
		MSGTMP2 = _mm_sha256msg1_epu32(MSGTMP2,MSGTMP3);

#define SHA256ROUND( \
msg, msgtmp0, msgtmp1, msgtmp2, msgtmp3, state0, state1, kvalue) \
	msg = msgtmp0; \
	msg = _mm_add_epi32(msg,_mm_load_si128(reinterpret_cast<const __m128i*>(kvalue))); \
	state1 = _mm_sha256rnds2_epu32(state1,state0,msg); \
	msgtmp1 = _mm_add_epi32(msgtmp1,_mm_alignr_epi8(msgtmp0,msgtmp3,4)); \
	msgtmp1 = _mm_sha256msg2_epu32(msgtmp1,msgtmp0); \
	msg = _mm_shuffle_epi32(msg,0x0E); \
	state0 = _mm_sha256rnds2_epu32(state0,state1,msg); \
	msgtmp3 = _mm_sha256msg1_epu32(msgtmp3,msgtmp0);

		//-- rounds 16-19, 20-23, 24-27, 28-31
		SHA256ROUND(MSG,MSGTMP0,MSGTMP1,MSGTMP2,MSGTMP3,STATE0,STATE1,&K64[16]);
		SHA256ROUND(MSG,MSGTMP1,MSGTMP2,MSGTMP3,MSGTMP0,STATE0,STATE1,&K64[20]);
		SHA256ROUND(MSG,MSGTMP2,MSGTMP3,MSGTMP0,MSGTMP1,STATE0,STATE1,&K64[24]);
		SHA256ROUND(MSG,MSGTMP3,MSGTMP0,MSGTMP1,MSGTMP2,STATE0,STATE1,&K64[28]);

		//-- rounds 32-35, 36-39, 40-43, 44-47
		SHA256ROUND(MSG,MSGTMP0,MSGTMP1,MSGTMP2,MSGTMP3,STATE0,STATE1,&K64[32]);
		SHA256ROUND(MSG,MSGTMP1,MSGTMP2,MSGTMP3,MSGTMP0,STATE0,STATE1,&K64[36]);
		SHA256ROUND(MSG,MSGTMP2,MSGTMP3,MSGTMP0,MSGTMP1,STATE0,STATE1,&K64[40]);
		SHA256ROUND(MSG,MSGTMP3,MSGTMP0,MSGTMP1,MSGTMP2,STATE0,STATE1,&K64[44]);

		//-- rounds 48-51
		SHA256ROUND(MSG,MSGTMP0,MSGTMP1,MSGTMP2,MSGTMP3,STATE0,STATE1,&K64[48]);

		// rounds 52-55
		MSG = MSGTMP1;
		MSG = _mm_add_epi32(MSG,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[52])));
		STATE1 = _mm_sha256rnds2_epu32(STATE1,STATE0,MSG);
		MSGTMP2 = _mm_add_epi32(MSGTMP2,_mm_alignr_epi8(MSGTMP1,MSGTMP0,4));
		MSGTMP2 = _mm_sha256msg2_epu32(MSGTMP2,MSGTMP1);
		MSG = _mm_shuffle_epi32(MSG,0x0E);
		STATE0 = _mm_sha256rnds2_epu32(STATE0,STATE1,MSG);

		// rounds 56-59
		MSG = MSGTMP2;
		MSG = _mm_add_epi32(MSG,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[56])));
		STATE1 = _mm_sha256rnds2_epu32(STATE1,STATE0,MSG);
		MSGTMP3 = _mm_add_epi32(MSGTMP3,_mm_alignr_epi8(MSGTMP2,MSGTMP1,4));
		MSGTMP3 = _mm_sha256msg2_epu32(MSGTMP3,MSGTMP2);
		MSG = _mm_shuffle_epi32(MSG,0x0E);
		STATE0 = _mm_sha256rnds2_epu32(STATE0,STATE1,MSG);

		// rounds 60-63
		MSG = MSGTMP3;
		MSG = _mm_add_epi32(MSG,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[60])));
		STATE1 = _mm_sha256rnds2_epu32(STATE1,STATE0,MSG);
		MSG = _mm_shuffle_epi32(MSG,0x0E);
		STATE0 = _mm_sha256rnds2_epu32(STATE0,STATE1,MSG);

		// add to state
		STATE0 = _mm_add_epi32(STATE0,ABEF_INIT);
		STATE1 = _mm_add_epi32(STATE1,CDGH_INIT);

		// reorder hash
		STATE0 = _mm_shuffle_epi32(STATE0,0x1B); // FEBA
		STATE1 = _mm_shuffle_epi32(STATE1,0xB1); // DCHG
		HASH0_SAVE = _mm_blend_epi16(STATE0,STATE1,0xF0); // DCBA
		HASH1_SAVE = _mm_alignr_epi8(STATE1,STATE0,8); // HGFE
	}

	// shuffle/return hash
	HASH0_SAVE = _mm_shuffle_epi8(HASH0_SAVE,SHUF_MASK);
	HASH1_SAVE = _mm_shuffle_epi8(HASH1_SAVE,SHUF_MASK);
	_mm_storeu_si128(reinterpret_cast<__m128i*>(&hash[0]),HASH0_SAVE);
	_mm_storeu_si128(reinterpret_cast<__m128i*>(&hash[16]),HASH1_SAVE);
}

void recursive_sha256_ni_x2(uint8_t* hash, const uint64_t num_iters)
{
	if(num_iters <= 0) {
		return;
	}

	alignas(64)
	static const uint32_t K64[64] = {
		0x428A2F98, 0x71374491, 0xB5C0FBCF, 0xE9B5DBA5,
		0x3956C25B, 0x59F111F1, 0x923F82A4, 0xAB1C5ED5,
		0xD807AA98, 0x12835B01, 0x243185BE, 0x550C7DC3,
		0x72BE5D74, 0x80DEB1FE, 0x9BDC06A7, 0xC19BF174,
		0xE49B69C1, 0xEFBE4786, 0x0FC19DC6, 0x240CA1CC,
		0x2DE92C6F, 0x4A7484AA, 0x5CB0A9DC, 0x76F988DA,
		0x983E5152, 0xA831C66D, 0xB00327C8, 0xBF597FC7,
		0xC6E00BF3, 0xD5A79147, 0x06CA6351, 0x14292967,
		0x27B70A85, 0x2E1B2138, 0x4D2C6DFC, 0x53380D13,
		0x650A7354, 0x766A0ABB, 0x81C2C92E, 0x92722C85,
		0xA2BFE8A1, 0xA81A664B, 0xC24B8B70, 0xC76C51A3,
		0xD192E819, 0xD6990624, 0xF40E3585, 0x106AA070,
		0x19A4C116, 0x1E376C08, 0x2748774C, 0x34B0BCB5,
		0x391C0CB3, 0x4ED8AA4A, 0x5B9CCA4F, 0x682E6FF3,
		0x748F82EE, 0x78A5636F, 0x84C87814, 0x8CC70208,
		0x90BEFFFA, 0xA4506CEB, 0xBEF9A3F7, 0xC67178F2
	};

	// shuffle mask
	const __m128i SHUF_MASK = _mm_set_epi64x(0x0c0d0e0f08090a0b,0x0405060700010203);

	// init values
	const __m128i ABEF_INIT = _mm_set_epi64x(0x6a09e667bb67ae85,0x510e527f9b05688c);
	const __m128i CDGH_INIT = _mm_set_epi64x(0x3c6ef372a54ff53a,0x1f83d9ab5be0cd19);

	// pre-calc/cache padding
	const __m128i HPAD0_CACHE = _mm_set_epi64x(0x0000000000000000,0x0000000080000000);
	const __m128i HPAD1_CACHE = _mm_set_epi64x(0x0000010000000000,0x0000000000000000);

	__m128i STATE0_P1;
	__m128i STATE1_P1;
	__m128i MSG_P1;
	__m128i MSGTMP0_P1;
	__m128i MSGTMP1_P1;
	__m128i MSGTMP2_P1;
	__m128i MSGTMP3_P1;
	__m128i STATE0_P2;
	__m128i STATE1_P2;
	__m128i MSG_P2;
	__m128i MSGTMP0_P2;
	__m128i MSGTMP1_P2;
	__m128i MSGTMP2_P2;
	__m128i MSGTMP3_P2;

	// init/shuffle hash
	__m128i HASH0_SAVE_P1 = _mm_loadu_si128(reinterpret_cast<__m128i*>(&hash[0]));
	__m128i HASH1_SAVE_P1 = _mm_loadu_si128(reinterpret_cast<__m128i*>(&hash[16]));
	__m128i HASH0_SAVE_P2 = _mm_loadu_si128(reinterpret_cast<__m128i*>(&hash[32]));
	__m128i HASH1_SAVE_P2 = _mm_loadu_si128(reinterpret_cast<__m128i*>(&hash[48]));
	HASH0_SAVE_P1 = _mm_shuffle_epi8(HASH0_SAVE_P1,SHUF_MASK);
	HASH1_SAVE_P1 = _mm_shuffle_epi8(HASH1_SAVE_P1,SHUF_MASK);
	HASH0_SAVE_P2 = _mm_shuffle_epi8(HASH0_SAVE_P2,SHUF_MASK);
	HASH1_SAVE_P2 = _mm_shuffle_epi8(HASH1_SAVE_P2,SHUF_MASK);

	for(uint64_t i = 0; i < num_iters; ++i) {

		// init state
		STATE0_P1 = ABEF_INIT;
		STATE1_P1 = CDGH_INIT;
		STATE0_P2 = ABEF_INIT;
		STATE1_P2 = CDGH_INIT;

		// rounds 0-3
		MSG_P1 = HASH0_SAVE_P1;
		MSG_P2 = HASH0_SAVE_P2;
		MSGTMP0_P1 = MSG_P1;
		MSGTMP0_P2 = MSG_P2;
		MSG_P1 = _mm_add_epi32(MSG_P1,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[0])));
		MSG_P2 = _mm_add_epi32(MSG_P2,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[0])));
		STATE1_P1 = _mm_sha256rnds2_epu32(STATE1_P1,STATE0_P1,MSG_P1);
		STATE1_P2 = _mm_sha256rnds2_epu32(STATE1_P2,STATE0_P2,MSG_P2);
		MSG_P1 = _mm_shuffle_epi32(MSG_P1,0x0E);
		MSG_P2 = _mm_shuffle_epi32(MSG_P2,0x0E);
		STATE0_P1 = _mm_sha256rnds2_epu32(STATE0_P1,STATE1_P1,MSG_P1);
		STATE0_P2 = _mm_sha256rnds2_epu32(STATE0_P2,STATE1_P2,MSG_P2);

		// rounds 4-7
		MSG_P1 = HASH1_SAVE_P1;
		MSG_P2 = HASH1_SAVE_P2;
		MSGTMP1_P1 = MSG_P1;
		MSGTMP1_P2 = MSG_P2;
		MSG_P1 = _mm_add_epi32(MSG_P1,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[4])));
		MSG_P2 = _mm_add_epi32(MSG_P2,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[4])));
		STATE1_P1 = _mm_sha256rnds2_epu32(STATE1_P1,STATE0_P1,MSG_P1);
		STATE1_P2 = _mm_sha256rnds2_epu32(STATE1_P2,STATE0_P2,MSG_P2);
		MSG_P1 = _mm_shuffle_epi32(MSG_P1,0x0E);
		MSG_P2 = _mm_shuffle_epi32(MSG_P2,0x0E);
		STATE0_P1 = _mm_sha256rnds2_epu32(STATE0_P1,STATE1_P1,MSG_P1);
		STATE0_P2 = _mm_sha256rnds2_epu32(STATE0_P2,STATE1_P2,MSG_P2);
		MSGTMP0_P1 = _mm_sha256msg1_epu32(MSGTMP0_P1,MSGTMP1_P1);
		MSGTMP0_P2 = _mm_sha256msg1_epu32(MSGTMP0_P2,MSGTMP1_P2);

		// rounds 8-11
		MSG_P1 = HPAD0_CACHE;
		MSG_P2 = HPAD0_CACHE;
		MSGTMP2_P1 = MSG_P1;
		MSGTMP2_P2 = MSG_P2;
		MSG_P1 = _mm_add_epi32(MSG_P1,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[8])));
		MSG_P2 = _mm_add_epi32(MSG_P2,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[8])));
		STATE1_P1 = _mm_sha256rnds2_epu32(STATE1_P1,STATE0_P1,MSG_P1);
		STATE1_P2 = _mm_sha256rnds2_epu32(STATE1_P2,STATE0_P2,MSG_P2);
		MSG_P1 = _mm_shuffle_epi32(MSG_P1,0x0E);
		MSG_P2 = _mm_shuffle_epi32(MSG_P2,0x0E);
		STATE0_P1 = _mm_sha256rnds2_epu32(STATE0_P1,STATE1_P1,MSG_P1);
		STATE0_P2 = _mm_sha256rnds2_epu32(STATE0_P2,STATE1_P2,MSG_P2);
		MSGTMP1_P1 = _mm_sha256msg1_epu32(MSGTMP1_P1,MSGTMP2_P1);
		MSGTMP1_P2 = _mm_sha256msg1_epu32(MSGTMP1_P2,MSGTMP2_P2);

		// rounds 12-15
		MSG_P1 = HPAD1_CACHE;
		MSG_P2 = HPAD1_CACHE;
		MSGTMP3_P1 = MSG_P1;
		MSGTMP3_P2 = MSG_P2;
		MSG_P1 = _mm_add_epi32(MSG_P1,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[12])));
		MSG_P2 = _mm_add_epi32(MSG_P2,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[12])));
		STATE1_P1 = _mm_sha256rnds2_epu32(STATE1_P1,STATE0_P1,MSG_P1);
		STATE1_P2 = _mm_sha256rnds2_epu32(STATE1_P2,STATE0_P2,MSG_P2);
		MSGTMP0_P1 = _mm_add_epi32(MSGTMP0_P1,_mm_alignr_epi8(MSGTMP3_P1,MSGTMP2_P1,4));
		MSGTMP0_P2 = _mm_add_epi32(MSGTMP0_P2,_mm_alignr_epi8(MSGTMP3_P2,MSGTMP2_P2,4));
		MSGTMP0_P1 = _mm_sha256msg2_epu32(MSGTMP0_P1,MSGTMP3_P1);
		MSGTMP0_P2 = _mm_sha256msg2_epu32(MSGTMP0_P2,MSGTMP3_P2);
		MSG_P1 = _mm_shuffle_epi32(MSG_P1,0x0E);
		MSG_P2 = _mm_shuffle_epi32(MSG_P2,0x0E);
		STATE0_P1 = _mm_sha256rnds2_epu32(STATE0_P1,STATE1_P1,MSG_P1);
		STATE0_P2 = _mm_sha256rnds2_epu32(STATE0_P2,STATE1_P2,MSG_P2);
		MSGTMP2_P1 = _mm_sha256msg1_epu32(MSGTMP2_P1,MSGTMP3_P1);
		MSGTMP2_P2 = _mm_sha256msg1_epu32(MSGTMP2_P2,MSGTMP3_P2);

#define SHA256ROUND_X2( \
msg_p1, msgtmp0_p1, msgtmp1_p1, msgtmp2_p1, msgtmp3_p1, state0_p1, state1_p1, \
msg_p2, msgtmp0_p2, msgtmp1_p2, msgtmp2_p2, msgtmp3_p2, state0_p2, state1_p2, kvalue) \
	msg_p1 = msgtmp0_p1; \
	msg_p2 = msgtmp0_p2; \
	msg_p1 = _mm_add_epi32(msg_p1,_mm_load_si128(reinterpret_cast<const __m128i*>(kvalue))); \
	msg_p2 = _mm_add_epi32(msg_p2,_mm_load_si128(reinterpret_cast<const __m128i*>(kvalue))); \
	state1_p1 = _mm_sha256rnds2_epu32(state1_p1,state0_p1,msg_p1); \
	state1_p2 = _mm_sha256rnds2_epu32(state1_p2,state0_p2,msg_p2); \
	msgtmp1_p1 = _mm_add_epi32(msgtmp1_p1,_mm_alignr_epi8(msgtmp0_p1,msgtmp3_p1,4)); \
	msgtmp1_p2 = _mm_add_epi32(msgtmp1_p2,_mm_alignr_epi8(msgtmp0_p2,msgtmp3_p2,4)); \
	msgtmp1_p1 = _mm_sha256msg2_epu32(msgtmp1_p1,msgtmp0_p1); \
	msgtmp1_p2 = _mm_sha256msg2_epu32(msgtmp1_p2,msgtmp0_p2); \
	msg_p1 = _mm_shuffle_epi32(msg_p1,0x0E); \
	msg_p2 = _mm_shuffle_epi32(msg_p2,0x0E); \
	state0_p1 = _mm_sha256rnds2_epu32(state0_p1,state1_p1,msg_p1); \
	state0_p2 = _mm_sha256rnds2_epu32(state0_p2,state1_p2,msg_p2); \
	msgtmp3_p1 = _mm_sha256msg1_epu32(msgtmp3_p1,msgtmp0_p1); \
	msgtmp3_p2 = _mm_sha256msg1_epu32(msgtmp3_p2,msgtmp0_p2);

		//-- rounds 16-19, 20-23, 24-27, 28-31
		SHA256ROUND_X2(MSG_P1,MSGTMP0_P1,MSGTMP1_P1,MSGTMP2_P1,MSGTMP3_P1,STATE0_P1,STATE1_P1,MSG_P2,MSGTMP0_P2,MSGTMP1_P2,MSGTMP2_P2,MSGTMP3_P2,STATE0_P2,STATE1_P2,&K64[16]);
		SHA256ROUND_X2(MSG_P1,MSGTMP1_P1,MSGTMP2_P1,MSGTMP3_P1,MSGTMP0_P1,STATE0_P1,STATE1_P1,MSG_P2,MSGTMP1_P2,MSGTMP2_P2,MSGTMP3_P2,MSGTMP0_P2,STATE0_P2,STATE1_P2,&K64[20]);
		SHA256ROUND_X2(MSG_P1,MSGTMP2_P1,MSGTMP3_P1,MSGTMP0_P1,MSGTMP1_P1,STATE0_P1,STATE1_P1,MSG_P2,MSGTMP2_P2,MSGTMP3_P2,MSGTMP0_P2,MSGTMP1_P2,STATE0_P2,STATE1_P2,&K64[24]);
		SHA256ROUND_X2(MSG_P1,MSGTMP3_P1,MSGTMP0_P1,MSGTMP1_P1,MSGTMP2_P1,STATE0_P1,STATE1_P1,MSG_P2,MSGTMP3_P2,MSGTMP0_P2,MSGTMP1_P2,MSGTMP2_P2,STATE0_P2,STATE1_P2,&K64[28]);

		//-- rounds 32-35, 36-39, 40-43, 44-47
		SHA256ROUND_X2(MSG_P1,MSGTMP0_P1,MSGTMP1_P1,MSGTMP2_P1,MSGTMP3_P1,STATE0_P1,STATE1_P1,MSG_P2,MSGTMP0_P2,MSGTMP1_P2,MSGTMP2_P2,MSGTMP3_P2,STATE0_P2,STATE1_P2,&K64[32]);
		SHA256ROUND_X2(MSG_P1,MSGTMP1_P1,MSGTMP2_P1,MSGTMP3_P1,MSGTMP0_P1,STATE0_P1,STATE1_P1,MSG_P2,MSGTMP1_P2,MSGTMP2_P2,MSGTMP3_P2,MSGTMP0_P2,STATE0_P2,STATE1_P2,&K64[36]);
		SHA256ROUND_X2(MSG_P1,MSGTMP2_P1,MSGTMP3_P1,MSGTMP0_P1,MSGTMP1_P1,STATE0_P1,STATE1_P1,MSG_P2,MSGTMP2_P2,MSGTMP3_P2,MSGTMP0_P2,MSGTMP1_P2,STATE0_P2,STATE1_P2,&K64[40]);
		SHA256ROUND_X2(MSG_P1,MSGTMP3_P1,MSGTMP0_P1,MSGTMP1_P1,MSGTMP2_P1,STATE0_P1,STATE1_P1,MSG_P2,MSGTMP3_P2,MSGTMP0_P2,MSGTMP1_P2,MSGTMP2_P2,STATE0_P2,STATE1_P2,&K64[44]);

		//-- rounds 48-51
		SHA256ROUND_X2(MSG_P1,MSGTMP0_P1,MSGTMP1_P1,MSGTMP2_P1,MSGTMP3_P1,STATE0_P1,STATE1_P1,MSG_P2,MSGTMP0_P2,MSGTMP1_P2,MSGTMP2_P2,MSGTMP3_P2,STATE0_P2,STATE1_P2,&K64[48]);

		// rounds 52-55
		MSG_P1 = MSGTMP1_P1;
		MSG_P2 = MSGTMP1_P2;
		MSG_P1 = _mm_add_epi32(MSG_P1,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[52])));
		MSG_P2 = _mm_add_epi32(MSG_P2,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[52])));
		STATE1_P1 = _mm_sha256rnds2_epu32(STATE1_P1,STATE0_P1,MSG_P1);
		STATE1_P2 = _mm_sha256rnds2_epu32(STATE1_P2,STATE0_P2,MSG_P2);
		MSGTMP2_P1 = _mm_add_epi32(MSGTMP2_P1,_mm_alignr_epi8(MSGTMP1_P1,MSGTMP0_P1,4));
		MSGTMP2_P2 = _mm_add_epi32(MSGTMP2_P2,_mm_alignr_epi8(MSGTMP1_P2,MSGTMP0_P2,4));
		MSGTMP2_P1 = _mm_sha256msg2_epu32(MSGTMP2_P1,MSGTMP1_P1);
		MSGTMP2_P2 = _mm_sha256msg2_epu32(MSGTMP2_P2,MSGTMP1_P2);
		MSG_P1 = _mm_shuffle_epi32(MSG_P1,0x0E);
		MSG_P2 = _mm_shuffle_epi32(MSG_P2,0x0E);
		STATE0_P1 = _mm_sha256rnds2_epu32(STATE0_P1,STATE1_P1,MSG_P1);
		STATE0_P2 = _mm_sha256rnds2_epu32(STATE0_P2,STATE1_P2,MSG_P2);

		// rounds 56-59
		MSG_P1 = MSGTMP2_P1;
		MSG_P2 = MSGTMP2_P2;
		MSG_P1 = _mm_add_epi32(MSG_P1,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[56])));
		MSG_P2 = _mm_add_epi32(MSG_P2,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[56])));
		STATE1_P1 = _mm_sha256rnds2_epu32(STATE1_P1,STATE0_P1,MSG_P1);
		STATE1_P2 = _mm_sha256rnds2_epu32(STATE1_P2,STATE0_P2,MSG_P2);
		MSGTMP3_P1 = _mm_add_epi32(MSGTMP3_P1,_mm_alignr_epi8(MSGTMP2_P1,MSGTMP1_P1,4));
		MSGTMP3_P2 = _mm_add_epi32(MSGTMP3_P2,_mm_alignr_epi8(MSGTMP2_P2,MSGTMP1_P2,4));
		MSGTMP3_P1 = _mm_sha256msg2_epu32(MSGTMP3_P1,MSGTMP2_P1);
		MSGTMP3_P2 = _mm_sha256msg2_epu32(MSGTMP3_P2,MSGTMP2_P2);
		MSG_P1 = _mm_shuffle_epi32(MSG_P1,0x0E);
		MSG_P2 = _mm_shuffle_epi32(MSG_P2,0x0E);
		STATE0_P1 = _mm_sha256rnds2_epu32(STATE0_P1,STATE1_P1,MSG_P1);
		STATE0_P2 = _mm_sha256rnds2_epu32(STATE0_P2,STATE1_P2,MSG_P2);

		// rounds 60-63
		MSG_P1 = MSGTMP3_P1;
		MSG_P2 = MSGTMP3_P2;
		MSG_P1 = _mm_add_epi32(MSG_P1,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[60])));
		MSG_P2 = _mm_add_epi32(MSG_P2,_mm_load_si128(reinterpret_cast<const __m128i*>(&K64[60])));
		STATE1_P1 = _mm_sha256rnds2_epu32(STATE1_P1,STATE0_P1,MSG_P1);
		STATE1_P2 = _mm_sha256rnds2_epu32(STATE1_P2,STATE0_P2,MSG_P2);
		MSG_P1 = _mm_shuffle_epi32(MSG_P1,0x0E);
		MSG_P2 = _mm_shuffle_epi32(MSG_P2,0x0E);
		STATE0_P1 = _mm_sha256rnds2_epu32(STATE0_P1,STATE1_P1,MSG_P1);
		STATE0_P2 = _mm_sha256rnds2_epu32(STATE0_P2,STATE1_P2,MSG_P2);

		// add to state
		STATE0_P1 = _mm_add_epi32(STATE0_P1,ABEF_INIT);
		STATE0_P2 = _mm_add_epi32(STATE0_P2,ABEF_INIT);
		STATE1_P1 = _mm_add_epi32(STATE1_P1,CDGH_INIT);
		STATE1_P2 = _mm_add_epi32(STATE1_P2,CDGH_INIT);

		// reorder hash
		STATE0_P1 = _mm_shuffle_epi32(STATE0_P1,0x1B); // FEBA
		STATE1_P1 = _mm_shuffle_epi32(STATE1_P1,0xB1); // DCHG
		STATE0_P2 = _mm_shuffle_epi32(STATE0_P2,0x1B); // FEBA
		STATE1_P2 = _mm_shuffle_epi32(STATE1_P2,0xB1); // DCHG
		HASH0_SAVE_P1 = _mm_blend_epi16(STATE0_P1,STATE1_P1,0xF0); // DCBA
		HASH1_SAVE_P1 = _mm_alignr_epi8(STATE1_P1,STATE0_P1,8); // HGFE
		HASH0_SAVE_P2 = _mm_blend_epi16(STATE0_P2,STATE1_P2,0xF0); // DCBA
		HASH1_SAVE_P2 = _mm_alignr_epi8(STATE1_P2,STATE0_P2,8); // HGFE
	}

	// shuffle/return hash
	HASH0_SAVE_P1 = _mm_shuffle_epi8(HASH0_SAVE_P1,SHUF_MASK);
	HASH1_SAVE_P1 = _mm_shuffle_epi8(HASH1_SAVE_P1,SHUF_MASK);
	HASH0_SAVE_P2 = _mm_shuffle_epi8(HASH0_SAVE_P2,SHUF_MASK);
	HASH1_SAVE_P2 = _mm_shuffle_epi8(HASH1_SAVE_P2,SHUF_MASK);
	_mm_storeu_si128(reinterpret_cast<__m128i*>(&hash[0]),HASH0_SAVE_P1);
	_mm_storeu_si128(reinterpret_cast<__m128i*>(&hash[16]),HASH1_SAVE_P1);
	_mm_storeu_si128(reinterpret_cast<__m128i*>(&hash[32]),HASH0_SAVE_P2);
	_mm_storeu_si128(reinterpret_cast<__m128i*>(&hash[48]),HASH1_SAVE_P2);
}

#else // __SHA__

void recursive_sha256_ni(uint8_t* hash, const uint64_t num_iters) {
	throw std::logic_error("recursive_sha256_ni() not available");
}

void recursive_sha256_ni_x2(uint8_t* hash, const uint64_t num_iters) {
	throw std::logic_error("recursive_sha256_ni_x2() not available");
}

#endif // __SHA__