encoder/basisu_uastc_enc.cpp

// basisu_uastc_enc.cpp
// Copyright (C) 2019-2021 Binomial LLC. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//    http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "basisu_uastc_enc.h"

#if BASISU_USE_ASTC_DECOMPRESS
#include "basisu_astc_decomp.h"
#endif

#include "basisu_gpu_texture.h"
#include "basisu_bc7enc.h"

#ifdef _DEBUG
// When BASISU_VALIDATE_UASTC_ENC is 1, we pack and unpack to/from UASTC and ASTC, then validate that each codec returns the exact same results. This is slower.
#define BASISU_VALIDATE_UASTC_ENC 1
#endif

#define BASISU_SUPPORT_FORCE_MODE 0

using namespace basist;

namespace basisu
{
	const uint32_t MAX_ENCODE_RESULTS = 512;

#if BASISU_VALIDATE_UASTC_ENC
	static void validate_func(bool condition, int line)
	{
		if (!condition)
		{
			fprintf(stderr, "basisu_uastc_enc: Internal validation failed on line %u!\n", line);
		}
	}

	#define VALIDATE(c) validate_func(c, __LINE__);
#else
	#define VALIDATE(c)
#endif

	enum dxt_constants
	{
		cDXT1SelectorBits = 2U, cDXT1SelectorValues = 1U << cDXT1SelectorBits, cDXT1SelectorMask = cDXT1SelectorValues - 1U,
		cDXT5SelectorBits = 3U, cDXT5SelectorValues = 1U << cDXT5SelectorBits, cDXT5SelectorMask = cDXT5SelectorValues - 1U,
	};

	struct dxt1_block
	{
		enum { cTotalEndpointBytes = 2, cTotalSelectorBytes = 4 };

		uint8_t m_low_color[cTotalEndpointBytes];
		uint8_t m_high_color[cTotalEndpointBytes];
		uint8_t m_selectors[cTotalSelectorBytes];

		inline void clear() { basisu::clear_obj(*this); }

		inline uint32_t get_high_color() const { return m_high_color[0] | (m_high_color[1] << 8U); }
		inline uint32_t get_low_color() const { return m_low_color[0] | (m_low_color[1] << 8U); }
		inline void set_low_color(uint16_t c) { m_low_color[0] = static_cast<uint8_t>(c & 0xFF); m_low_color[1] = static_cast<uint8_t>((c >> 8) & 0xFF); }
		inline void set_high_color(uint16_t c) { m_high_color[0] = static_cast<uint8_t>(c & 0xFF); m_high_color[1] = static_cast<uint8_t>((c >> 8) & 0xFF); }
		inline uint32_t get_selector(uint32_t x, uint32_t y) const { assert((x < 4U) && (y < 4U)); return (m_selectors[y] >> (x * cDXT1SelectorBits))& cDXT1SelectorMask; }
		inline void set_selector(uint32_t x, uint32_t y, uint32_t val) { assert((x < 4U) && (y < 4U) && (val < 4U)); m_selectors[y] &= (~(cDXT1SelectorMask << (x * cDXT1SelectorBits))); m_selectors[y] |= (val << (x * cDXT1SelectorBits)); }

		static uint16_t pack_color(const color_rgba& color, bool scaled, uint32_t bias = 127U)
		{
			uint32_t r = color.r, g = color.g, b = color.b;
			if (scaled)
			{
				r = (r * 31U + bias) / 255U;
				g = (g * 63U + bias) / 255U;
				b = (b * 31U + bias) / 255U;
			}
			return static_cast<uint16_t>(basisu::minimum(b, 31U) | (basisu::minimum(g, 63U) << 5U) | (basisu::minimum(r, 31U) << 11U));
		}

		static uint16_t pack_unscaled_color(uint32_t r, uint32_t g, uint32_t b) { return static_cast<uint16_t>(b | (g << 5U) | (r << 11U)); }
	};

#define UASTC_WRITE_MODE_DESCS 0

	static inline void uastc_write_bits(uint8_t* pBuf, uint32_t& bit_offset, uint64_t code, uint32_t codesize, const char* pDesc)
	{
		(void)pDesc;

#if UASTC_WRITE_MODE_DESCS
		if (pDesc)
			printf("%s: %u %u\n", pDesc, bit_offset, codesize);
#endif

		assert((codesize == 64) || (code < (1ULL << codesize)));

		while (codesize)
		{
			uint32_t byte_bit_offset = bit_offset & 7;
			uint32_t bits_to_write = basisu::minimum<int>(codesize, 8 - byte_bit_offset);

			pBuf[bit_offset >> 3] |= (code << byte_bit_offset);

			code >>= bits_to_write;
			codesize -= bits_to_write;
			bit_offset += bits_to_write;
		}
	}

	void pack_uastc(basist::uastc_block& blk, const uastc_encode_results& result, const etc_block& etc1_blk, uint32_t etc1_bias, const eac_a8_block& etc_eac_a8_blk, bool bc1_hint0, bool bc1_hint1)
	{
		if ((g_uastc_mode_has_alpha[result.m_uastc_mode]) && (result.m_uastc_mode != UASTC_MODE_INDEX_SOLID_COLOR))
		{
			assert(etc_eac_a8_blk.m_multiplier >= 1);
		}

		uint8_t buf[32];
		memset(buf, 0, sizeof(buf));

		uint32_t block_bit_offset = 0;

#if UASTC_WRITE_MODE_DESCS
		printf("**** Mode: %u\n", result.m_uastc_mode);
#endif

		uastc_write_bits(buf, block_bit_offset, g_uastc_mode_huff_codes[result.m_uastc_mode][0], g_uastc_mode_huff_codes[result.m_uastc_mode][1], "mode");

		if (result.m_uastc_mode == UASTC_MODE_INDEX_SOLID_COLOR)
		{
			uastc_write_bits(buf, block_bit_offset, result.m_solid_color.r, 8, "R");
			uastc_write_bits(buf, block_bit_offset, result.m_solid_color.g, 8, "G");
			uastc_write_bits(buf, block_bit_offset, result.m_solid_color.b, 8, "B");
			uastc_write_bits(buf, block_bit_offset, result.m_solid_color.a, 8, "A");

			uastc_write_bits(buf, block_bit_offset, etc1_blk.get_diff_bit(), 1, "ETC1D");
			uastc_write_bits(buf, block_bit_offset, etc1_blk.get_inten_table(0), 3, "ETC1I");
			uastc_write_bits(buf, block_bit_offset, etc1_blk.get_selector(0, 0), 2, "ETC1S");

			uint32_t r, g, b;
			if (etc1_blk.get_diff_bit())
				etc_block::unpack_color5(r, g, b, etc1_blk.get_base5_color(), false);
			else
				etc_block::unpack_color4(r, g, b, etc1_blk.get_base4_color(0), false);

			uastc_write_bits(buf, block_bit_offset, r, 5, "ETC1R");
			uastc_write_bits(buf, block_bit_offset, g, 5, "ETC1G");
			uastc_write_bits(buf, block_bit_offset, b, 5, "ETC1B");

			memcpy(&blk, buf, sizeof(blk));
			return;
		}

		if (g_uastc_mode_has_bc1_hint0[result.m_uastc_mode])
			uastc_write_bits(buf, block_bit_offset, bc1_hint0, 1, "BC1H0");
		else
		{
			assert(bc1_hint0 == false);
		}

		if (g_uastc_mode_has_bc1_hint1[result.m_uastc_mode])
			uastc_write_bits(buf, block_bit_offset, bc1_hint1, 1, "BC1H1");
		else
		{
			assert(bc1_hint1 == false);
		}

		uastc_write_bits(buf, block_bit_offset, etc1_blk.get_flip_bit(), 1, "ETC1F");
		uastc_write_bits(buf, block_bit_offset, etc1_blk.get_diff_bit(), 1, "ETC1D");
		uastc_write_bits(buf, block_bit_offset, etc1_blk.get_inten_table(0), 3, "ETC1I0");
		uastc_write_bits(buf, block_bit_offset, etc1_blk.get_inten_table(1), 3, "ETC1I1");

		if (g_uastc_mode_has_etc1_bias[result.m_uastc_mode])
			uastc_write_bits(buf, block_bit_offset, etc1_bias, 5, "ETC1BIAS");
		else
		{
			assert(etc1_bias == 0);
		}

		if (g_uastc_mode_has_alpha[result.m_uastc_mode])
		{
			const uint32_t etc2_hints = etc_eac_a8_blk.m_table | (etc_eac_a8_blk.m_multiplier << 4);

			assert(etc2_hints > 0 && etc2_hints <= 0xFF);
			uastc_write_bits(buf, block_bit_offset, etc2_hints, 8, "ETC2TM");
		}

		uint32_t subsets = 1;
		switch (result.m_uastc_mode)
		{
		case 2:
		case 4:
		case 7:
		case 9:
		case 16:
			uastc_write_bits(buf, block_bit_offset, result.m_common_pattern, 5, "PAT");
			subsets = 2;
			break;
		case 3:
			uastc_write_bits(buf, block_bit_offset, result.m_common_pattern, 4, "PAT");
			subsets = 3;
			break;
		default:
			break;
		}

#ifdef _DEBUG
		uint32_t part_seed = 0;
		switch (result.m_uastc_mode)
		{
		case 2:
		case 4:
		case 9:
		case 16:
			part_seed = g_astc_bc7_common_partitions2[result.m_common_pattern].m_astc;
			break;
		case 3:
			part_seed = g_astc_bc7_common_partitions3[result.m_common_pattern].m_astc;
			break;
		case 7:
			part_seed = g_bc7_3_astc2_common_partitions[result.m_common_pattern].m_astc2;
			break;
		default:
			break;
		}
#endif		

		uint32_t total_planes = 1;
		switch (result.m_uastc_mode)
		{
		case 6:
		case 11:
		case 13:
			uastc_write_bits(buf, block_bit_offset, result.m_astc.m_ccs, 2, "COMPSEL");
			total_planes = 2;
			break;
		case 17:
			// CCS field is always 3 for dual plane LA.
			assert(result.m_astc.m_ccs == 3);
			total_planes = 2;
			break;
		default:
			break;
		}

		uint8_t weights[32];
		memcpy(weights, result.m_astc.m_weights, 16 * total_planes);

		uint8_t endpoints[18];
		memcpy(endpoints, result.m_astc.m_endpoints, sizeof(endpoints));

		const uint32_t total_comps = g_uastc_mode_comps[result.m_uastc_mode];

		// LLAA
		// LLAA LLAA
		// LLAA LLAA LLAA
		// RRGGBB
		// RRGGBB RRGGBB
		// RRGGBB RRGGBB RRGGBB
		// RRGGBBAA
		// RRGGBBAA RRGGBBAA

		const uint32_t weight_bits = g_uastc_mode_weight_bits[result.m_uastc_mode];

		const uint8_t* pPartition_pattern;
		const uint8_t* pSubset_anchor_indices = basist::get_anchor_indices(subsets, result.m_uastc_mode, result.m_common_pattern, pPartition_pattern);

		for (uint32_t plane_index = 0; plane_index < total_planes; plane_index++)
		{
			for (uint32_t subset_index = 0; subset_index < subsets; subset_index++)
			{
				const uint32_t anchor_index = pSubset_anchor_indices[subset_index];

#ifdef _DEBUG
				if (subsets >= 2)
				{
					for (uint32_t i = 0; i < 16; i++)
					{
						const uint32_t part_index = astc_compute_texel_partition(part_seed, i & 3, i >> 2, 0, subsets, true);
						if (part_index == subset_index)
						{
							assert(anchor_index == i);
							break;
						}
					}
				}
				else
				{
					assert(!anchor_index);
				}
#endif

				// Check anchor weight's MSB - if it's set then invert this subset's weights and swap the endpoints
				if (weights[anchor_index * total_planes + plane_index] & (1 << (weight_bits - 1)))
				{
					for (uint32_t i = 0; i < 16; i++)
					{
						const uint32_t part_index = pPartition_pattern[i];

#ifdef _DEBUG
						if (subsets >= 2)
						{
							assert(part_index == (uint32_t)astc_compute_texel_partition(part_seed, i & 3, i >> 2, 0, subsets, true));
						}
						else
						{
							assert(!part_index);
						}
#endif

						if (part_index == subset_index)
							weights[i * total_planes + plane_index] = ((1 << weight_bits) - 1) - weights[i * total_planes + plane_index];
					}

					if (total_planes == 2)
					{
						for (int c = 0; c < (int)total_comps; c++)
						{
							const uint32_t comp_plane = (total_comps == 2) ? c : ((c == result.m_astc.m_ccs) ? 1 : 0);

							if (comp_plane == plane_index)
								std::swap(endpoints[c * 2 + 0], endpoints[c * 2 + 1]);
						}
					}
					else
					{
						for (uint32_t c = 0; c < total_comps; c++)
							std::swap(endpoints[subset_index * total_comps * 2 + c * 2 + 0], endpoints[subset_index * total_comps * 2 + c * 2 + 1]);
					}
				}
			} // subset_index
		} // plane_index

		const uint32_t total_values = total_comps * 2 * subsets;
		const uint32_t endpoint_range = g_uastc_mode_endpoint_ranges[result.m_uastc_mode];

		uint32_t bit_values[18];
		uint32_t tq_values[8];
		uint32_t total_tq_values = 0;
		uint32_t tq_accum = 0;
		uint32_t tq_mul = 1;

		const uint32_t ep_bits = g_astc_bise_range_table[endpoint_range][0];
		const uint32_t ep_trits = g_astc_bise_range_table[endpoint_range][1];
		const uint32_t ep_quints = g_astc_bise_range_table[endpoint_range][2];

		for (uint32_t i = 0; i < total_values; i++)
		{
			uint32_t val = endpoints[i];

			uint32_t bits = val & ((1 << ep_bits) - 1);
			uint32_t tq = val >> ep_bits;

			bit_values[i] = bits;

			if (ep_trits)
			{
				assert(tq < 3);
				tq_accum += tq * tq_mul;
				tq_mul *= 3;
				if (tq_mul == 243)
				{
					tq_values[total_tq_values++] = tq_accum;
					tq_accum = 0;
					tq_mul = 1;
				}
			}
			else if (ep_quints)
			{
				assert(tq < 5);
				tq_accum += tq * tq_mul;
				tq_mul *= 5;
				if (tq_mul == 125)
				{
					tq_values[total_tq_values++] = tq_accum;
					tq_accum = 0;
					tq_mul = 1;
				}
			}
		}

		uint32_t total_endpoint_bits = 0;

		for (uint32_t i = 0; i < total_tq_values; i++)
		{
			const uint32_t num_bits = ep_trits ? 8 : 7;
			uastc_write_bits(buf, block_bit_offset, tq_values[i], num_bits, "ETQ");
			total_endpoint_bits += num_bits;
		}

		if (tq_mul > 1)
		{
			uint32_t num_bits;
			if (ep_trits)
			{
				if (tq_mul == 3)
					num_bits = 2;
				else if (tq_mul == 9)
					num_bits = 4;
				else if (tq_mul == 27)
					num_bits = 5;
				else //if (tq_mul == 81)
					num_bits = 7;
			}
			else
			{
				if (tq_mul == 5)
					num_bits = 3;
				else //if (tq_mul == 25)
					num_bits = 5;
			}
			uastc_write_bits(buf, block_bit_offset, tq_accum, num_bits, "ETQ");
			total_endpoint_bits += num_bits;
		}

		for (uint32_t i = 0; i < total_values; i++)
		{
			uastc_write_bits(buf, block_bit_offset, bit_values[i], ep_bits, "EBITS");
			total_endpoint_bits += ep_bits;
		}

#if UASTC_WRITE_MODE_DESCS
		uint32_t weight_start = block_bit_offset;
#endif

		uint32_t total_weight_bits = 0;
		const uint32_t plane_shift = (total_planes == 2) ? 1 : 0;
		for (uint32_t i = 0; i < 16 * total_planes; i++)
		{
			uint32_t numbits = weight_bits;
			for (uint32_t s = 0; s < subsets; s++)
			{
				if (pSubset_anchor_indices[s] == (i >> plane_shift))
				{
					numbits--;
					break;
				}
			}

			uastc_write_bits(buf, block_bit_offset, weights[i], numbits, nullptr);

			total_weight_bits += numbits;
		}

#if UASTC_WRITE_MODE_DESCS
		printf("WEIGHTS: %u %u\n", weight_start, total_weight_bits);
#endif

		assert(block_bit_offset <= 128);
		memcpy(&blk, buf, sizeof(blk));

#if UASTC_WRITE_MODE_DESCS
		printf("Total bits: %u, endpoint bits: %u, weight bits: %u\n", block_bit_offset, total_endpoint_bits, total_weight_bits);
#endif
	}
	
	// MODE 0
	// 0. DualPlane: 0, WeightRange: 8 (16), Subsets: 1, CEM: 8 (RGB Direct       ), EndpointRange: 19 (192)       MODE6 RGB
	// 18. DualPlane: 0, WeightRange: 11 (32), Subsets: 1, CEM: 8 (RGB Direct       ), EndpointRange: 11 (32)       MODE6 RGB
	static void astc_mode0_or_18(uint32_t mode, const color_rgba block[4][4], uastc_encode_results* pResults, uint32_t& total_results, bc7enc_compress_block_params& comp_params, const uint8_t *pForce_selectors = nullptr)
	{
		const uint32_t endpoint_range = (mode == 18) ? 11 : 19;
		const uint32_t weight_range = (mode == 18) ? 11 : 8;

		color_cell_compressor_params ccell_params;
		memset(&ccell_params, 0, sizeof(ccell_params));

		ccell_params.m_num_pixels = 16;
		ccell_params.m_pPixels = (color_quad_u8*)&block[0][0];
		ccell_params.m_num_selector_weights = (mode == 18) ? 32 : 16;
		ccell_params.m_pSelector_weights = (mode == 18) ? g_astc_weights5 : g_astc_weights4;
		ccell_params.m_pSelector_weightsx = (mode == 18) ? (const bc7enc_vec4F*)g_astc_weights5x : (const bc7enc_vec4F*)g_astc_weights4x;
		ccell_params.m_astc_endpoint_range = endpoint_range;
		ccell_params.m_weights[0] = 1;
		ccell_params.m_weights[1] = 1;
		ccell_params.m_weights[2] = 1;
		ccell_params.m_weights[3] = 1;
		ccell_params.m_pForce_selectors = pForce_selectors;

		color_cell_compressor_results ccell_results;
		uint8_t ccell_result_selectors[16];
		uint8_t ccell_result_selectors_temp[16];
		memset(&ccell_results, 0, sizeof(ccell_results));
		ccell_results.m_pSelectors = &ccell_result_selectors[0];
		ccell_results.m_pSelectors_temp = &ccell_result_selectors_temp[0];

		uint64_t part_err = color_cell_compression(255, &ccell_params, &ccell_results, &comp_params);

		// ASTC
		astc_block_desc astc_results;
		memset(&astc_results, 0, sizeof(astc_results));

		astc_results.m_dual_plane = false;
		astc_results.m_weight_range = weight_range;// (mode == 18) ? 11 : 8;

		astc_results.m_ccs = 0;
		astc_results.m_subsets = 1;
		astc_results.m_partition_seed = 0;
		astc_results.m_cem = 8;

		astc_results.m_endpoints[0] = ccell_results.m_astc_low_endpoint.m_c[0];
		astc_results.m_endpoints[1] = ccell_results.m_astc_high_endpoint.m_c[0];
		astc_results.m_endpoints[2] = ccell_results.m_astc_low_endpoint.m_c[1];
		astc_results.m_endpoints[3] = ccell_results.m_astc_high_endpoint.m_c[1];
		astc_results.m_endpoints[4] = ccell_results.m_astc_low_endpoint.m_c[2];
		astc_results.m_endpoints[5] = ccell_results.m_astc_high_endpoint.m_c[2];
				
		bool invert = false;

		if (pForce_selectors == nullptr)
		{
			int s0 = g_astc_unquant[endpoint_range][astc_results.m_endpoints[0]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[2]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[4]].m_unquant;
			int s1 = g_astc_unquant[endpoint_range][astc_results.m_endpoints[1]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[3]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[5]].m_unquant;
			if (s1 < s0)
			{
				std::swap(astc_results.m_endpoints[0], astc_results.m_endpoints[1]);
				std::swap(astc_results.m_endpoints[2], astc_results.m_endpoints[3]);
				std::swap(astc_results.m_endpoints[4], astc_results.m_endpoints[5]);
				invert = true;
			}
		}

		for (uint32_t y = 0; y < 4; y++)
		{
			for (uint32_t x = 0; x < 4; x++)
			{
				astc_results.m_weights[x + y * 4] = ccell_result_selectors[x + y * 4];

				if (invert)
					astc_results.m_weights[x + y * 4] = ((mode == 18) ? 31 : 15) - astc_results.m_weights[x + y * 4];
			}
		}

		assert(total_results < MAX_ENCODE_RESULTS);
		if (total_results < MAX_ENCODE_RESULTS)
		{
			pResults[total_results].m_uastc_mode = mode;
			pResults[total_results].m_common_pattern = 0;
			pResults[total_results].m_astc = astc_results;
			pResults[total_results].m_astc_err = part_err;
			total_results++;
		}
	}

	// MODE 1
	// 1-subset, 2-bit indices, 8-bit endpoints, BC7 mode 3
	// DualPlane: 0, WeightRange: 2 (4), Subsets: 1, CEM: 8 (RGB Direct       ), EndpointRange: 20 (256)        MODE3 or MODE5 RGB
	static void astc_mode1(const color_rgba block[4][4], uastc_encode_results* pResults, uint32_t& total_results, bc7enc_compress_block_params& comp_params)
	{
		color_cell_compressor_params ccell_params;
		memset(&ccell_params, 0, sizeof(ccell_params));

		ccell_params.m_num_pixels = 16;
		ccell_params.m_pPixels = (color_quad_u8*)&block[0][0];
		ccell_params.m_num_selector_weights = 4;
		ccell_params.m_pSelector_weights = g_bc7_weights2;
		ccell_params.m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights2x;
		ccell_params.m_astc_endpoint_range = 20;
		ccell_params.m_weights[0] = 1;
		ccell_params.m_weights[1] = 1;
		ccell_params.m_weights[2] = 1;
		ccell_params.m_weights[3] = 1;

		color_cell_compressor_results ccell_results;
		uint8_t ccell_result_selectors[16];
		uint8_t ccell_result_selectors_temp[16];
		memset(&ccell_results, 0, sizeof(ccell_results));
		ccell_results.m_pSelectors = &ccell_result_selectors[0];
		ccell_results.m_pSelectors_temp = &ccell_result_selectors_temp[0];

		uint64_t part_err = color_cell_compression(255, &ccell_params, &ccell_results, &comp_params);

		// ASTC
		astc_block_desc astc_results;
		memset(&astc_results, 0, sizeof(astc_results));

		astc_results.m_dual_plane = false;
		astc_results.m_weight_range = 2;

		astc_results.m_ccs = 0;
		astc_results.m_subsets = 1;
		astc_results.m_partition_seed = 0;
		astc_results.m_cem = 8;

		astc_results.m_endpoints[0] = ccell_results.m_astc_low_endpoint.m_c[0];
		astc_results.m_endpoints[1] = ccell_results.m_astc_high_endpoint.m_c[0];
		astc_results.m_endpoints[2] = ccell_results.m_astc_low_endpoint.m_c[1];
		astc_results.m_endpoints[3] = ccell_results.m_astc_high_endpoint.m_c[1];
		astc_results.m_endpoints[4] = ccell_results.m_astc_low_endpoint.m_c[2];
		astc_results.m_endpoints[5] = ccell_results.m_astc_high_endpoint.m_c[2];

		const uint32_t range = 20;

		bool invert = false;

		int s0 = g_astc_unquant[range][astc_results.m_endpoints[0]].m_unquant + g_astc_unquant[range][astc_results.m_endpoints[2]].m_unquant + g_astc_unquant[range][astc_results.m_endpoints[4]].m_unquant;
		int s1 = g_astc_unquant[range][astc_results.m_endpoints[1]].m_unquant + g_astc_unquant[range][astc_results.m_endpoints[3]].m_unquant + g_astc_unquant[range][astc_results.m_endpoints[5]].m_unquant;
		if (s1 < s0)
		{
			std::swap(astc_results.m_endpoints[0], astc_results.m_endpoints[1]);
			std::swap(astc_results.m_endpoints[2], astc_results.m_endpoints[3]);
			std::swap(astc_results.m_endpoints[4], astc_results.m_endpoints[5]);
			invert = true;
		}

		for (uint32_t y = 0; y < 4; y++)
		{
			for (uint32_t x = 0; x < 4; x++)
			{
				astc_results.m_weights[x + y * 4] = ccell_result_selectors[x + y * 4];

				if (invert)
					astc_results.m_weights[x + y * 4] = 3 - astc_results.m_weights[x + y * 4];
			}
		}

		assert(total_results < MAX_ENCODE_RESULTS);
		if (total_results < MAX_ENCODE_RESULTS)
		{
			pResults[total_results].m_uastc_mode = 1;
			pResults[total_results].m_common_pattern = 0;
			pResults[total_results].m_astc = astc_results;
			pResults[total_results].m_astc_err = part_err;
			total_results++;
		}
	}

	static uint32_t estimate_partition2(uint32_t num_weights, uint32_t num_comps, const uint32_t* pWeights, const color_rgba block[4][4], const uint32_t weights[4])
	{
		assert(pWeights[0] == 0 && pWeights[num_weights - 1] == 64);

		uint64_t best_err = UINT64_MAX;
		uint32_t best_common_pattern = 0;

		for (uint32_t common_pattern = 0; common_pattern < TOTAL_ASTC_BC7_COMMON_PARTITIONS2; common_pattern++)
		{
			const uint32_t bc7_pattern = g_astc_bc7_common_partitions2[common_pattern].m_bc7;

			const uint8_t* pPartition = &g_bc7_partition2[bc7_pattern * 16];

			color_quad_u8 subset_colors[2][16];
			uint32_t subset_total_colors[2] = { 0, 0 };
			for (uint32_t index = 0; index < 16; index++)
				subset_colors[pPartition[index]][subset_total_colors[pPartition[index]]++] = ((const color_quad_u8*)block)[index];

			uint64_t total_subset_err = 0;
			for (uint32_t subset = 0; (subset < 2) && (total_subset_err < best_err); subset++)
				total_subset_err += color_cell_compression_est_astc(num_weights, num_comps, pWeights, subset_total_colors[subset], &subset_colors[subset][0], best_err, weights);

			if (total_subset_err < best_err)
			{
				best_err = total_subset_err;
				best_common_pattern = common_pattern;
			}
		}

		return best_common_pattern;
	}

	// MODE 2
	// 2-subset, 3-bit indices, 4-bit endpoints, BC7 mode 1
	// DualPlane: 0, WeightRange: 5 (8), Subsets: 2, CEM: 8 (RGB Direct       ), EndpointRange: 8 (16)          MODE1
	static void astc_mode2(const color_rgba block[4][4], uastc_encode_results* pResults, uint32_t& total_results, bc7enc_compress_block_params& comp_params, bool estimate_partition)
	{
		uint32_t first_common_pattern = 0;
		uint32_t last_common_pattern = TOTAL_ASTC_BC7_COMMON_PARTITIONS2;

		if (estimate_partition)
		{
			const uint32_t weights[4] = { 1, 1, 1, 1 };
			first_common_pattern = estimate_partition2(8, 3, g_bc7_weights3, block, weights);
			last_common_pattern = first_common_pattern + 1;
		}

		for (uint32_t common_pattern = first_common_pattern; common_pattern < last_common_pattern; common_pattern++)
		{
			const uint32_t bc7_pattern = g_astc_bc7_common_partitions2[common_pattern].m_bc7;

			color_rgba part_pixels[2][16];
			uint32_t part_pixel_index[4][4];
			uint32_t num_part_pixels[2] = { 0, 0 };

			for (uint32_t y = 0; y < 4; y++)
			{
				for (uint32_t x = 0; x < 4; x++)
				{
					const uint32_t part = g_bc7_partition2[16 * bc7_pattern + x + y * 4];
					part_pixel_index[y][x] = num_part_pixels[part];
					part_pixels[part][num_part_pixels[part]++] = block[y][x];
				}
			}

			color_cell_compressor_params ccell_params[2];
			color_cell_compressor_results ccell_results[2];
			uint8_t ccell_result_selectors[2][16];
			uint8_t ccell_result_selectors_temp[2][16];

			uint64_t total_part_err = 0;
			for (uint32_t part = 0; part < 2; part++)
			{
				memset(&ccell_params[part], 0, sizeof(ccell_params[part]));

				ccell_params[part].m_num_pixels = num_part_pixels[part];
				ccell_params[part].m_pPixels = (color_quad_u8*)&part_pixels[part][0];
				ccell_params[part].m_num_selector_weights = 8;
				ccell_params[part].m_pSelector_weights = g_bc7_weights3;
				ccell_params[part].m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights3x;
				ccell_params[part].m_astc_endpoint_range = 8;
				ccell_params[part].m_weights[0] = 1;
				ccell_params[part].m_weights[1] = 1;
				ccell_params[part].m_weights[2] = 1;
				ccell_params[part].m_weights[3] = 1;

				memset(&ccell_results[part], 0, sizeof(ccell_results[part]));
				ccell_results[part].m_pSelectors = &ccell_result_selectors[part][0];
				ccell_results[part].m_pSelectors_temp = &ccell_result_selectors_temp[part][0];

				uint64_t part_err = color_cell_compression(255, &ccell_params[part], &ccell_results[part], &comp_params);
				total_part_err += part_err;
			} // part

			{
				// ASTC
				astc_block_desc astc_results;
				memset(&astc_results, 0, sizeof(astc_results));

				astc_results.m_dual_plane = false;
				astc_results.m_weight_range = 5;

				astc_results.m_ccs = 0;
				astc_results.m_subsets = 2;
				astc_results.m_partition_seed = g_astc_bc7_common_partitions2[common_pattern].m_astc;
				astc_results.m_cem = 8;

				uint32_t p0 = 0;
				uint32_t p1 = 1;
				if (g_astc_bc7_common_partitions2[common_pattern].m_invert)
					std::swap(p0, p1);

				astc_results.m_endpoints[0] = ccell_results[p0].m_astc_low_endpoint.m_c[0];
				astc_results.m_endpoints[1] = ccell_results[p0].m_astc_high_endpoint.m_c[0];
				astc_results.m_endpoints[2] = ccell_results[p0].m_astc_low_endpoint.m_c[1];
				astc_results.m_endpoints[3] = ccell_results[p0].m_astc_high_endpoint.m_c[1];
				astc_results.m_endpoints[4] = ccell_results[p0].m_astc_low_endpoint.m_c[2];
				astc_results.m_endpoints[5] = ccell_results[p0].m_astc_high_endpoint.m_c[2];

				const uint32_t range = 8;

				bool invert[2] = { false, false };

				int s0 = g_astc_unquant[range][astc_results.m_endpoints[0]].m_unquant + g_astc_unquant[range][astc_results.m_endpoints[2]].m_unquant + g_astc_unquant[range][astc_results.m_endpoints[4]].m_unquant;
				int s1 = g_astc_unquant[range][astc_results.m_endpoints[1]].m_unquant + g_astc_unquant[range][astc_results.m_endpoints[3]].m_unquant + g_astc_unquant[range][astc_results.m_endpoints[5]].m_unquant;
				if (s1 < s0)
				{
					std::swap(astc_results.m_endpoints[0], astc_results.m_endpoints[1]);
					std::swap(astc_results.m_endpoints[2], astc_results.m_endpoints[3]);
					std::swap(astc_results.m_endpoints[4], astc_results.m_endpoints[5]);
					invert[0] = true;
				}

				astc_results.m_endpoints[6] = ccell_results[p1].m_astc_low_endpoint.m_c[0];
				astc_results.m_endpoints[7] = ccell_results[p1].m_astc_high_endpoint.m_c[0];
				astc_results.m_endpoints[8] = ccell_results[p1].m_astc_low_endpoint.m_c[1];
				astc_results.m_endpoints[9] = ccell_results[p1].m_astc_high_endpoint.m_c[1];
				astc_results.m_endpoints[10] = ccell_results[p1].m_astc_low_endpoint.m_c[2];
				astc_results.m_endpoints[11] = ccell_results[p1].m_astc_high_endpoint.m_c[2];

				s0 = g_astc_unquant[range][astc_results.m_endpoints[0 + 6]].m_unquant + g_astc_unquant[range][astc_results.m_endpoints[2 + 6]].m_unquant + g_astc_unquant[range][astc_results.m_endpoints[4 + 6]].m_unquant;
				s1 = g_astc_unquant[range][astc_results.m_endpoints[1 + 6]].m_unquant + g_astc_unquant[range][astc_results.m_endpoints[3 + 6]].m_unquant + g_astc_unquant[range][astc_results.m_endpoints[5 + 6]].m_unquant;

				if (s1 < s0)
				{
					std::swap(astc_results.m_endpoints[0 + 6], astc_results.m_endpoints[1 + 6]);
					std::swap(astc_results.m_endpoints[2 + 6], astc_results.m_endpoints[3 + 6]);
					std::swap(astc_results.m_endpoints[4 + 6], astc_results.m_endpoints[5 + 6]);
					invert[1] = true;
				}

				for (uint32_t y = 0; y < 4; y++)
				{
					for (uint32_t x = 0; x < 4; x++)
					{
						const uint32_t bc7_part = g_bc7_partition2[16 * bc7_pattern + x + y * 4];

						astc_results.m_weights[x + y * 4] = ccell_result_selectors[bc7_part][part_pixel_index[y][x]];

						uint32_t astc_part = bc7_part;
						if (g_astc_bc7_common_partitions2[common_pattern].m_invert)
							astc_part = 1 - astc_part;

						if (invert[astc_part])
							astc_results.m_weights[x + y * 4] = 7 - astc_results.m_weights[x + y * 4];
					}
				}

				assert(total_results < MAX_ENCODE_RESULTS);
				if (total_results < MAX_ENCODE_RESULTS)
				{
					pResults[total_results].m_uastc_mode = 2;
					pResults[total_results].m_common_pattern = common_pattern;
					pResults[total_results].m_astc = astc_results;
					pResults[total_results].m_astc_err = total_part_err;
					total_results++;
				}
			}

		} // common_pattern
	}

	// MODE 3
	// 3-subsets, 2-bit indices, [0,11] endpoints, BC7 mode 2
	// DualPlane: 0, WeightRange: 2 (4), Subsets: 3, CEM: 8 (RGB Direct	     ), EndpointRange: 7 (12)		   MODE2
	static void astc_mode3(const color_rgba block[4][4], uastc_encode_results* pResults, uint32_t& total_results, bc7enc_compress_block_params& comp_params, bool estimate_partition)
	{
		uint32_t first_common_pattern = 0;
		uint32_t last_common_pattern = TOTAL_ASTC_BC7_COMMON_PARTITIONS3;

		if (estimate_partition)
		{
			uint64_t best_err = UINT64_MAX;
			uint32_t best_common_pattern = 0;
			const uint32_t weights[4] = { 1, 1, 1, 1 };

			for (uint32_t common_pattern = 0; common_pattern < TOTAL_ASTC_BC7_COMMON_PARTITIONS3; common_pattern++)
			{
				const uint32_t bc7_pattern = g_astc_bc7_common_partitions3[common_pattern].m_bc7;

				const uint8_t* pPartition = &g_bc7_partition3[bc7_pattern * 16];

				color_quad_u8 subset_colors[3][16];
				uint32_t subset_total_colors[3] = { 0, 0 };
				for (uint32_t index = 0; index < 16; index++)
					subset_colors[pPartition[index]][subset_total_colors[pPartition[index]]++] = ((const color_quad_u8*)block)[index];

				uint64_t total_subset_err = 0;
				for (uint32_t subset = 0; (subset < 3) && (total_subset_err < best_err); subset++)
					total_subset_err += color_cell_compression_est_astc(4, 3, g_bc7_weights2, subset_total_colors[subset], &subset_colors[subset][0], best_err, weights);

				if (total_subset_err < best_err)
				{
					best_err = total_subset_err;
					best_common_pattern = common_pattern;
				}
			}

			first_common_pattern = best_common_pattern;
			last_common_pattern = best_common_pattern + 1;
		}

		for (uint32_t common_pattern = first_common_pattern; common_pattern < last_common_pattern; common_pattern++)
		{
			const uint32_t endpoint_range = 7;

			const uint32_t bc7_pattern = g_astc_bc7_common_partitions3[common_pattern].m_bc7;

			color_rgba part_pixels[3][16];
			uint32_t part_pixel_index[4][4];
			uint32_t num_part_pixels[3] = { 0, 0, 0 };

			for (uint32_t y = 0; y < 4; y++)
			{
				for (uint32_t x = 0; x < 4; x++)
				{
					const uint32_t bc7_part = g_bc7_partition3[16 * bc7_pattern + x + y * 4];
					part_pixel_index[y][x] = num_part_pixels[bc7_part];
					part_pixels[bc7_part][num_part_pixels[bc7_part]++] = block[y][x];
				}
			}

			color_cell_compressor_params ccell_params[3];
			color_cell_compressor_results ccell_results[3];
			uint8_t ccell_result_selectors[3][16];
			uint8_t ccell_result_selectors_temp[3][16];

			uint64_t total_part_err = 0;
			for (uint32_t bc7_part = 0; bc7_part < 3; bc7_part++)
			{
				memset(&ccell_params[bc7_part], 0, sizeof(ccell_params[bc7_part]));

				ccell_params[bc7_part].m_num_pixels = num_part_pixels[bc7_part];
				ccell_params[bc7_part].m_pPixels = (color_quad_u8*)&part_pixels[bc7_part][0];
				ccell_params[bc7_part].m_num_selector_weights = 4;
				ccell_params[bc7_part].m_pSelector_weights = g_bc7_weights2;
				ccell_params[bc7_part].m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights2x;
				ccell_params[bc7_part].m_astc_endpoint_range = endpoint_range;
				ccell_params[bc7_part].m_weights[0] = 1;
				ccell_params[bc7_part].m_weights[1] = 1;
				ccell_params[bc7_part].m_weights[2] = 1;
				ccell_params[bc7_part].m_weights[3] = 1;

				memset(&ccell_results[bc7_part], 0, sizeof(ccell_results[bc7_part]));
				ccell_results[bc7_part].m_pSelectors = &ccell_result_selectors[bc7_part][0];
				ccell_results[bc7_part].m_pSelectors_temp = &ccell_result_selectors_temp[bc7_part][0];

				uint64_t part_err = color_cell_compression(255, &ccell_params[bc7_part], &ccell_results[bc7_part], &comp_params);
				total_part_err += part_err;
			} // part

			{
				// ASTC
				astc_block_desc astc_results;
				memset(&astc_results, 0, sizeof(astc_results));

				astc_results.m_dual_plane = false;
				astc_results.m_weight_range = 2;

				astc_results.m_ccs = 0;
				astc_results.m_subsets = 3;
				astc_results.m_partition_seed = g_astc_bc7_common_partitions3[common_pattern].m_astc;
				astc_results.m_cem = 8;

				uint32_t astc_to_bc7_part[3]; // converts ASTC to BC7 partition index
				const uint32_t perm = g_astc_bc7_common_partitions3[common_pattern].m_astc_to_bc7_perm;
				astc_to_bc7_part[0] = g_astc_to_bc7_partition_index_perm_tables[perm][0];
				astc_to_bc7_part[1] = g_astc_to_bc7_partition_index_perm_tables[perm][1];
				astc_to_bc7_part[2] = g_astc_to_bc7_partition_index_perm_tables[perm][2];

				bool invert_astc_part[3] = { false, false, false };

				for (uint32_t astc_part = 0; astc_part < 3; astc_part++)
				{
					uint8_t* pEndpoints = &astc_results.m_endpoints[6 * astc_part];

					pEndpoints[0] = ccell_results[astc_to_bc7_part[astc_part]].m_astc_low_endpoint.m_c[0];
					pEndpoints[1] = ccell_results[astc_to_bc7_part[astc_part]].m_astc_high_endpoint.m_c[0];
					pEndpoints[2] = ccell_results[astc_to_bc7_part[astc_part]].m_astc_low_endpoint.m_c[1];
					pEndpoints[3] = ccell_results[astc_to_bc7_part[astc_part]].m_astc_high_endpoint.m_c[1];
					pEndpoints[4] = ccell_results[astc_to_bc7_part[astc_part]].m_astc_low_endpoint.m_c[2];
					pEndpoints[5] = ccell_results[astc_to_bc7_part[astc_part]].m_astc_high_endpoint.m_c[2];

					int s0 = g_astc_unquant[endpoint_range][pEndpoints[0]].m_unquant + g_astc_unquant[endpoint_range][pEndpoints[2]].m_unquant + g_astc_unquant[endpoint_range][pEndpoints[4]].m_unquant;
					int s1 = g_astc_unquant[endpoint_range][pEndpoints[1]].m_unquant + g_astc_unquant[endpoint_range][pEndpoints[3]].m_unquant + g_astc_unquant[endpoint_range][pEndpoints[5]].m_unquant;
					if (s1 < s0)
					{
						std::swap(pEndpoints[0], pEndpoints[1]);
						std::swap(pEndpoints[2], pEndpoints[3]);
						std::swap(pEndpoints[4], pEndpoints[5]);
						invert_astc_part[astc_part] = true;
					}
				}

				for (uint32_t y = 0; y < 4; y++)
				{
					for (uint32_t x = 0; x < 4; x++)
					{
						const uint32_t bc7_part = g_bc7_partition3[16 * bc7_pattern + x + y * 4];

						astc_results.m_weights[x + y * 4] = ccell_result_selectors[bc7_part][part_pixel_index[y][x]];

						uint32_t astc_part = 0;
						for (uint32_t i = 0; i < 3; i++)
						{
							if (astc_to_bc7_part[i] == bc7_part)
							{
								astc_part = i;
								break;
							}
						}

						if (invert_astc_part[astc_part])
							astc_results.m_weights[x + y * 4] = 3 - astc_results.m_weights[x + y * 4];
					}
				}

				assert(total_results < MAX_ENCODE_RESULTS);
				if (total_results < MAX_ENCODE_RESULTS)
				{
					pResults[total_results].m_uastc_mode = 3;
					pResults[total_results].m_common_pattern = common_pattern;
					pResults[total_results].m_astc = astc_results;
					pResults[total_results].m_astc_err = total_part_err;
					total_results++;
				}

			}

		} // common_pattern
	}

	// MODE 4
	// DualPlane: 0, WeightRange: 2 (4), Subsets: 2, CEM: 8 (RGB Direct       ), EndpointRange: 12 (40)         MODE3
	static void astc_mode4(const color_rgba block[4][4], uastc_encode_results* pResults, uint32_t& total_results, bc7enc_compress_block_params& comp_params, bool estimate_partition)
	{
		//const uint32_t weight_range = 2;
		const uint32_t endpoint_range = 12;

		uint32_t first_common_pattern = 0;
		uint32_t last_common_pattern = TOTAL_ASTC_BC7_COMMON_PARTITIONS2;

		if (estimate_partition)
		{
			const uint32_t weights[4] = { 1, 1, 1, 1 };
			first_common_pattern = estimate_partition2(4, 3, g_bc7_weights2, block, weights);
			last_common_pattern = first_common_pattern + 1;
		}

		for (uint32_t common_pattern = first_common_pattern; common_pattern < last_common_pattern; common_pattern++)
		{
			const uint32_t bc7_pattern = g_astc_bc7_common_partitions2[common_pattern].m_bc7;

			color_rgba part_pixels[2][16];
			uint32_t part_pixel_index[4][4];
			uint32_t num_part_pixels[2] = { 0, 0 };

			for (uint32_t y = 0; y < 4; y++)
			{
				for (uint32_t x = 0; x < 4; x++)
				{
					const uint32_t part = g_bc7_partition2[16 * bc7_pattern + x + y * 4];
					part_pixel_index[y][x] = num_part_pixels[part];
					part_pixels[part][num_part_pixels[part]++] = block[y][x];
				}
			}

			color_cell_compressor_params ccell_params[2];
			color_cell_compressor_results ccell_results[2];
			uint8_t ccell_result_selectors[2][16];
			uint8_t ccell_result_selectors_temp[2][16];

			uint64_t total_part_err = 0;
			for (uint32_t part = 0; part < 2; part++)
			{
				memset(&ccell_params[part], 0, sizeof(ccell_params[part]));

				ccell_params[part].m_num_pixels = num_part_pixels[part];
				ccell_params[part].m_pPixels = (color_quad_u8*)&part_pixels[part][0];
				ccell_params[part].m_num_selector_weights = 4;
				ccell_params[part].m_pSelector_weights = g_bc7_weights2;
				ccell_params[part].m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights2x;
				ccell_params[part].m_astc_endpoint_range = endpoint_range;
				ccell_params[part].m_weights[0] = 1;
				ccell_params[part].m_weights[1] = 1;
				ccell_params[part].m_weights[2] = 1;
				ccell_params[part].m_weights[3] = 1;

				memset(&ccell_results[part], 0, sizeof(ccell_results[part]));
				ccell_results[part].m_pSelectors = &ccell_result_selectors[part][0];
				ccell_results[part].m_pSelectors_temp = &ccell_result_selectors_temp[part][0];

				uint64_t part_err = color_cell_compression(255, &ccell_params[part], &ccell_results[part], &comp_params);
				total_part_err += part_err;
			} // part

			// ASTC
			astc_block_desc astc_results;
			memset(&astc_results, 0, sizeof(astc_results));

			astc_results.m_dual_plane = false;
			astc_results.m_weight_range = 2;

			astc_results.m_ccs = 0;
			astc_results.m_subsets = 2;
			astc_results.m_partition_seed = g_astc_bc7_common_partitions2[common_pattern].m_astc;
			astc_results.m_cem = 8;

			uint32_t p0 = 0;
			uint32_t p1 = 1;
			if (g_astc_bc7_common_partitions2[common_pattern].m_invert)
				std::swap(p0, p1);

			astc_results.m_endpoints[0] = ccell_results[p0].m_astc_low_endpoint.m_c[0];
			astc_results.m_endpoints[1] = ccell_results[p0].m_astc_high_endpoint.m_c[0];
			astc_results.m_endpoints[2] = ccell_results[p0].m_astc_low_endpoint.m_c[1];
			astc_results.m_endpoints[3] = ccell_results[p0].m_astc_high_endpoint.m_c[1];
			astc_results.m_endpoints[4] = ccell_results[p0].m_astc_low_endpoint.m_c[2];
			astc_results.m_endpoints[5] = ccell_results[p0].m_astc_high_endpoint.m_c[2];

			bool invert[2] = { false, false };

			int s0 = g_astc_unquant[endpoint_range][astc_results.m_endpoints[0]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[2]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[4]].m_unquant;
			int s1 = g_astc_unquant[endpoint_range][astc_results.m_endpoints[1]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[3]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[5]].m_unquant;
			if (s1 < s0)
			{
				std::swap(astc_results.m_endpoints[0], astc_results.m_endpoints[1]);
				std::swap(astc_results.m_endpoints[2], astc_results.m_endpoints[3]);
				std::swap(astc_results.m_endpoints[4], astc_results.m_endpoints[5]);
				invert[0] = true;
			}

			astc_results.m_endpoints[6] = ccell_results[p1].m_astc_low_endpoint.m_c[0];
			astc_results.m_endpoints[7] = ccell_results[p1].m_astc_high_endpoint.m_c[0];
			astc_results.m_endpoints[8] = ccell_results[p1].m_astc_low_endpoint.m_c[1];
			astc_results.m_endpoints[9] = ccell_results[p1].m_astc_high_endpoint.m_c[1];
			astc_results.m_endpoints[10] = ccell_results[p1].m_astc_low_endpoint.m_c[2];
			astc_results.m_endpoints[11] = ccell_results[p1].m_astc_high_endpoint.m_c[2];

			s0 = g_astc_unquant[endpoint_range][astc_results.m_endpoints[0 + 6]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[2 + 6]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[4 + 6]].m_unquant;
			s1 = g_astc_unquant[endpoint_range][astc_results.m_endpoints[1 + 6]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[3 + 6]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[5 + 6]].m_unquant;

			if (s1 < s0)
			{
				std::swap(astc_results.m_endpoints[0 + 6], astc_results.m_endpoints[1 + 6]);
				std::swap(astc_results.m_endpoints[2 + 6], astc_results.m_endpoints[3 + 6]);
				std::swap(astc_results.m_endpoints[4 + 6], astc_results.m_endpoints[5 + 6]);
				invert[1] = true;
			}

			for (uint32_t y = 0; y < 4; y++)
			{
				for (uint32_t x = 0; x < 4; x++)
				{
					const uint32_t bc7_part = g_bc7_partition2[16 * bc7_pattern + x + y * 4];

					astc_results.m_weights[x + y * 4] = ccell_result_selectors[bc7_part][part_pixel_index[y][x]];

					uint32_t astc_part = bc7_part;
					if (g_astc_bc7_common_partitions2[common_pattern].m_invert)
						astc_part = 1 - astc_part;

					if (invert[astc_part])
						astc_results.m_weights[x + y * 4] = 3 - astc_results.m_weights[x + y * 4];
				}
			}

			assert(total_results < MAX_ENCODE_RESULTS);
			if (total_results < MAX_ENCODE_RESULTS)
			{
				pResults[total_results].m_uastc_mode = 4;
				pResults[total_results].m_common_pattern = common_pattern;
				pResults[total_results].m_astc = astc_results;
				pResults[total_results].m_astc_err = total_part_err;
				total_results++;
			}

		} // common_pattern
	}

	// MODE 5 
	// DualPlane: 0, WeightRange: 5 (8), Subsets: 1, CEM: 8 (RGB Direct       ), EndpointRange: 20 (256) 		BC7 MODE 6 (or MODE 1 1-subset)
	static void astc_mode5(const color_rgba block[4][4], uastc_encode_results* pResults, uint32_t& total_results, bc7enc_compress_block_params& comp_params)
	{
		const uint32_t weight_range = 5;
		const uint32_t endpoint_range = 20;

		color_cell_compressor_params ccell_params;
		memset(&ccell_params, 0, sizeof(ccell_params));

		ccell_params.m_num_pixels = 16;
		ccell_params.m_pPixels = (color_quad_u8*)&block[0][0];
		ccell_params.m_num_selector_weights = 8;
		ccell_params.m_pSelector_weights = g_bc7_weights3;
		ccell_params.m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights3x;
		ccell_params.m_astc_endpoint_range = endpoint_range;
		ccell_params.m_weights[0] = 1;
		ccell_params.m_weights[1] = 1;
		ccell_params.m_weights[2] = 1;
		ccell_params.m_weights[3] = 1;

		color_cell_compressor_results ccell_results;
		uint8_t ccell_result_selectors[16];
		uint8_t ccell_result_selectors_temp[16];
		memset(&ccell_results, 0, sizeof(ccell_results));
		ccell_results.m_pSelectors = &ccell_result_selectors[0];
		ccell_results.m_pSelectors_temp = &ccell_result_selectors_temp[0];

		uint64_t part_err = color_cell_compression(255, &ccell_params, &ccell_results, &comp_params);

		// ASTC
		astc_block_desc blk;
		memset(&blk, 0, sizeof(blk));

		blk.m_dual_plane = false;
		blk.m_weight_range = weight_range;

		blk.m_ccs = 0;
		blk.m_subsets = 1;
		blk.m_partition_seed = 0;
		blk.m_cem = 8;

		blk.m_endpoints[0] = ccell_results.m_astc_low_endpoint.m_c[0];
		blk.m_endpoints[1] = ccell_results.m_astc_high_endpoint.m_c[0];
		blk.m_endpoints[2] = ccell_results.m_astc_low_endpoint.m_c[1];
		blk.m_endpoints[3] = ccell_results.m_astc_high_endpoint.m_c[1];
		blk.m_endpoints[4] = ccell_results.m_astc_low_endpoint.m_c[2];
		blk.m_endpoints[5] = ccell_results.m_astc_high_endpoint.m_c[2];

		bool invert = false;

		int s0 = g_astc_unquant[endpoint_range][blk.m_endpoints[0]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[2]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[4]].m_unquant;
		int s1 = g_astc_unquant[endpoint_range][blk.m_endpoints[1]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[3]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[5]].m_unquant;
		if (s1 < s0)
		{
			std::swap(blk.m_endpoints[0], blk.m_endpoints[1]);
			std::swap(blk.m_endpoints[2], blk.m_endpoints[3]);
			std::swap(blk.m_endpoints[4], blk.m_endpoints[5]);
			invert = true;
		}

		for (uint32_t y = 0; y < 4; y++)
		{
			for (uint32_t x = 0; x < 4; x++)
			{
				blk.m_weights[x + y * 4] = ccell_result_selectors[x + y * 4];

				if (invert)
					blk.m_weights[x + y * 4] = 7 - blk.m_weights[x + y * 4];
			}
		}

		assert(total_results < MAX_ENCODE_RESULTS);
		if (total_results < MAX_ENCODE_RESULTS)
		{
			pResults[total_results].m_uastc_mode = 5;
			pResults[total_results].m_common_pattern = 0;
			pResults[total_results].m_astc = blk;
			pResults[total_results].m_astc_err = part_err;
			total_results++;
		}
	}

	// MODE 6
	// DualPlane: 1, WeightRange: 2 (4), Subsets: 1, CEM: 8 (RGB Direct       ), EndpointRange: 18 (160)		BC7 MODE5
	static void astc_mode6(const color_rgba block[4][4], uastc_encode_results* pResults, uint32_t& total_results, bc7enc_compress_block_params& comp_params)
	{
		for (uint32_t rot_comp = 0; rot_comp < 3; rot_comp++)
		{
			const uint32_t weight_range = 2;
			const uint32_t endpoint_range = 18;

			color_quad_u8 block_rgb[16];
			color_quad_u8 block_a[16];
			for (uint32_t i = 0; i < 16; i++)
			{
				block_rgb[i] = ((color_quad_u8*)&block[0][0])[i];
				block_a[i] = block_rgb[i];

				uint8_t c = block_a[i].m_c[rot_comp];
				block_a[i].m_c[0] = c;
				block_a[i].m_c[1] = c;
				block_a[i].m_c[2] = c;
				block_a[i].m_c[3] = 255;

				block_rgb[i].m_c[rot_comp] = 255;
			}

			uint8_t ccell_result_selectors_temp[16];

			color_cell_compressor_params ccell_params_rgb;
			memset(&ccell_params_rgb, 0, sizeof(ccell_params_rgb));

			ccell_params_rgb.m_num_pixels = 16;
			ccell_params_rgb.m_pPixels = block_rgb;
			ccell_params_rgb.m_num_selector_weights = 4;
			ccell_params_rgb.m_pSelector_weights = g_bc7_weights2;
			ccell_params_rgb.m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights2x;
			ccell_params_rgb.m_astc_endpoint_range = endpoint_range;
			ccell_params_rgb.m_weights[0] = 1;
			ccell_params_rgb.m_weights[1] = 1;
			ccell_params_rgb.m_weights[2] = 1;
			ccell_params_rgb.m_weights[3] = 1;

			color_cell_compressor_results ccell_results_rgb;
			uint8_t ccell_result_selectors_rgb[16];
			memset(&ccell_results_rgb, 0, sizeof(ccell_results_rgb));
			ccell_results_rgb.m_pSelectors = &ccell_result_selectors_rgb[0];
			ccell_results_rgb.m_pSelectors_temp = &ccell_result_selectors_temp[0];

			uint64_t part_err_rgb = color_cell_compression(255, &ccell_params_rgb, &ccell_results_rgb, &comp_params);
			
			color_cell_compressor_params ccell_params_a;
			memset(&ccell_params_a, 0, sizeof(ccell_params_a));

			ccell_params_a.m_num_pixels = 16;
			ccell_params_a.m_pPixels = block_a;
			ccell_params_a.m_num_selector_weights = 4;
			ccell_params_a.m_pSelector_weights = g_bc7_weights2;
			ccell_params_a.m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights2x;
			ccell_params_a.m_astc_endpoint_range = endpoint_range;
			ccell_params_a.m_weights[0] = 1;
			ccell_params_a.m_weights[1] = 1;
			ccell_params_a.m_weights[2] = 1;
			ccell_params_a.m_weights[3] = 1;

			color_cell_compressor_results ccell_results_a;
			uint8_t ccell_result_selectors_a[16];
			memset(&ccell_results_a, 0, sizeof(ccell_results_a));
			ccell_results_a.m_pSelectors = &ccell_result_selectors_a[0];
			ccell_results_a.m_pSelectors_temp = &ccell_result_selectors_temp[0];

			uint64_t part_err_a = color_cell_compression(255, &ccell_params_a, &ccell_results_a, &comp_params) / 3;

			uint64_t total_err = part_err_rgb + part_err_a;

			// ASTC
			astc_block_desc blk;
			memset(&blk, 0, sizeof(blk));

			blk.m_dual_plane = true;
			blk.m_weight_range = weight_range;

			blk.m_ccs = rot_comp;
			blk.m_subsets = 1;
			blk.m_partition_seed = 0;
			blk.m_cem = 8;

			blk.m_endpoints[0] = (rot_comp == 0 ? ccell_results_a : ccell_results_rgb).m_astc_low_endpoint.m_c[0];
			blk.m_endpoints[1] = (rot_comp == 0 ? ccell_results_a : ccell_results_rgb).m_astc_high_endpoint.m_c[0];
			blk.m_endpoints[2] = (rot_comp == 1 ? ccell_results_a : ccell_results_rgb).m_astc_low_endpoint.m_c[1];
			blk.m_endpoints[3] = (rot_comp == 1 ? ccell_results_a : ccell_results_rgb).m_astc_high_endpoint.m_c[1];
			blk.m_endpoints[4] = (rot_comp == 2 ? ccell_results_a : ccell_results_rgb).m_astc_low_endpoint.m_c[2];
			blk.m_endpoints[5] = (rot_comp == 2 ? ccell_results_a : ccell_results_rgb).m_astc_high_endpoint.m_c[2];

			bool invert = false;

			int s0 = g_astc_unquant[endpoint_range][blk.m_endpoints[0]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[2]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[4]].m_unquant;
			int s1 = g_astc_unquant[endpoint_range][blk.m_endpoints[1]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[3]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[5]].m_unquant;
			if (s1 < s0)
			{
				std::swap(blk.m_endpoints[0], blk.m_endpoints[1]);
				std::swap(blk.m_endpoints[2], blk.m_endpoints[3]);
				std::swap(blk.m_endpoints[4], blk.m_endpoints[5]);
				invert = true;
			}

			for (uint32_t y = 0; y < 4; y++)
			{
				for (uint32_t x = 0; x < 4; x++)
				{
					uint32_t rgb_index = ccell_result_selectors_rgb[x + y * 4];
					uint32_t a_index = ccell_result_selectors_a[x + y * 4];

					if (invert)
					{
						rgb_index = 3 - rgb_index;
						a_index = 3 - a_index;
					}

					blk.m_weights[(x + y * 4) * 2 + 0] = (uint8_t)rgb_index;
					blk.m_weights[(x + y * 4) * 2 + 1] = (uint8_t)a_index;
				}
			}

			assert(total_results < MAX_ENCODE_RESULTS);
			if (total_results < MAX_ENCODE_RESULTS)
			{
				pResults[total_results].m_uastc_mode = 6;
				pResults[total_results].m_common_pattern = 0;
				pResults[total_results].m_astc = blk;
				pResults[total_results].m_astc_err = total_err;
				total_results++;
			}
		} // rot_comp
	}

	// MODE 7 - 2 subset ASTC, 3 subset BC7
	// DualPlane: 0, WeightRange: 2 (4), Subsets: 2, CEM: 8 (RGB Direct       ), EndpointRange: 12 (40)         MODE2
	static void astc_mode7(const color_rgba block[4][4], uastc_encode_results* pResults, uint32_t& total_results, bc7enc_compress_block_params& comp_params, bool estimate_partition)
	{
		uint32_t first_common_pattern = 0;
		uint32_t last_common_pattern = TOTAL_BC7_3_ASTC2_COMMON_PARTITIONS;

		if (estimate_partition)
		{
			uint64_t best_err = UINT64_MAX;
			uint32_t best_common_pattern = 0;
			const uint32_t weights[4] = { 1, 1, 1, 1 };

			for (uint32_t common_pattern = 0; common_pattern < TOTAL_BC7_3_ASTC2_COMMON_PARTITIONS; common_pattern++)
			{
				const uint8_t* pPartition = &g_bc7_3_astc2_patterns2[common_pattern][0];

#ifdef _DEBUG
				const uint32_t astc_pattern = g_bc7_3_astc2_common_partitions[common_pattern].m_astc2;
				const uint32_t bc7_pattern = g_bc7_3_astc2_common_partitions[common_pattern].m_bc73;
				const uint32_t common_pattern_k = g_bc7_3_astc2_common_partitions[common_pattern].k;

				for (uint32_t y = 0; y < 4; y++)
				{
					for (uint32_t x = 0; x < 4; x++)
					{
						const uint32_t astc_part = bc7_convert_partition_index_3_to_2(g_bc7_partition3[16 * bc7_pattern + x + y * 4], common_pattern_k);
						assert((int)astc_part == astc_compute_texel_partition(astc_pattern, x, y, 0, 2, true));
						assert(astc_part == pPartition[x + y * 4]);
					}
				}
#endif

				color_quad_u8 subset_colors[2][16];
				uint32_t subset_total_colors[2] = { 0, 0 };
				for (uint32_t index = 0; index < 16; index++)
					subset_colors[pPartition[index]][subset_total_colors[pPartition[index]]++] = ((const color_quad_u8*)block)[index];

				uint64_t total_subset_err = 0;
				for (uint32_t subset = 0; (subset < 2) && (total_subset_err < best_err); subset++)
					total_subset_err += color_cell_compression_est_astc(4, 3, g_bc7_weights2, subset_total_colors[subset], &subset_colors[subset][0], best_err, weights);

				if (total_subset_err < best_err)
				{
					best_err = total_subset_err;
					best_common_pattern = common_pattern;
				}
			}

			first_common_pattern = best_common_pattern;
			last_common_pattern = best_common_pattern + 1;
		}

		//const uint32_t weight_range = 2;
		const uint32_t endpoint_range = 12;

		for (uint32_t common_pattern = first_common_pattern; common_pattern < last_common_pattern; common_pattern++)
		{
			const uint32_t astc_pattern = g_bc7_3_astc2_common_partitions[common_pattern].m_astc2;
			const uint32_t bc7_pattern = g_bc7_3_astc2_common_partitions[common_pattern].m_bc73;
			const uint32_t common_pattern_k = g_bc7_3_astc2_common_partitions[common_pattern].k;

			color_rgba part_pixels[2][16];
			uint32_t part_pixel_index[4][4];
			uint32_t num_part_pixels[2] = { 0, 0 };

			for (uint32_t y = 0; y < 4; y++)
			{
				for (uint32_t x = 0; x < 4; x++)
				{
					const uint32_t astc_part = bc7_convert_partition_index_3_to_2(g_bc7_partition3[16 * bc7_pattern + x + y * 4], common_pattern_k);
#ifdef _DEBUG					
					assert((int)astc_part == astc_compute_texel_partition(astc_pattern, x, y, 0, 2, true));
#endif					

					part_pixel_index[y][x] = num_part_pixels[astc_part];
					part_pixels[astc_part][num_part_pixels[astc_part]++] = block[y][x];
				}
			}

			color_cell_compressor_params ccell_params[2];
			color_cell_compressor_results ccell_results[2];
			uint8_t ccell_result_selectors[2][16];
			uint8_t ccell_result_selectors_temp[2][16];

			uint64_t total_part_err = 0;
			for (uint32_t part = 0; part < 2; part++)
			{
				memset(&ccell_params[part], 0, sizeof(ccell_params[part]));

				ccell_params[part].m_num_pixels = num_part_pixels[part];
				ccell_params[part].m_pPixels = (color_quad_u8*)&part_pixels[part][0];
				ccell_params[part].m_num_selector_weights = 4;
				ccell_params[part].m_pSelector_weights = g_bc7_weights2;
				ccell_params[part].m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights2x;
				ccell_params[part].m_astc_endpoint_range = endpoint_range;
				ccell_params[part].m_weights[0] = 1;
				ccell_params[part].m_weights[1] = 1;
				ccell_params[part].m_weights[2] = 1;
				ccell_params[part].m_weights[3] = 1;

				memset(&ccell_results[part], 0, sizeof(ccell_results[part]));
				ccell_results[part].m_pSelectors = &ccell_result_selectors[part][0];
				ccell_results[part].m_pSelectors_temp = &ccell_result_selectors_temp[part][0];

				uint64_t part_err = color_cell_compression(255, &ccell_params[part], &ccell_results[part], &comp_params);
				total_part_err += part_err;
			} // part

			// ASTC
			astc_block_desc blk;
			memset(&blk, 0, sizeof(blk));

			blk.m_dual_plane = false;
			blk.m_weight_range = 2;

			blk.m_ccs = 0;
			blk.m_subsets = 2;
			blk.m_partition_seed = astc_pattern;
			blk.m_cem = 8;

			const uint32_t p0 = 0;
			const uint32_t p1 = 1;

			blk.m_endpoints[0] = ccell_results[p0].m_astc_low_endpoint.m_c[0];
			blk.m_endpoints[1] = ccell_results[p0].m_astc_high_endpoint.m_c[0];
			blk.m_endpoints[2] = ccell_results[p0].m_astc_low_endpoint.m_c[1];
			blk.m_endpoints[3] = ccell_results[p0].m_astc_high_endpoint.m_c[1];
			blk.m_endpoints[4] = ccell_results[p0].m_astc_low_endpoint.m_c[2];
			blk.m_endpoints[5] = ccell_results[p0].m_astc_high_endpoint.m_c[2];

			bool invert[2] = { false, false };

			int s0 = g_astc_unquant[endpoint_range][blk.m_endpoints[0]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[2]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[4]].m_unquant;
			int s1 = g_astc_unquant[endpoint_range][blk.m_endpoints[1]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[3]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[5]].m_unquant;
			if (s1 < s0)
			{
				std::swap(blk.m_endpoints[0], blk.m_endpoints[1]);
				std::swap(blk.m_endpoints[2], blk.m_endpoints[3]);
				std::swap(blk.m_endpoints[4], blk.m_endpoints[5]);
				invert[0] = true;
			}

			blk.m_endpoints[6] = ccell_results[p1].m_astc_low_endpoint.m_c[0];
			blk.m_endpoints[7] = ccell_results[p1].m_astc_high_endpoint.m_c[0];
			blk.m_endpoints[8] = ccell_results[p1].m_astc_low_endpoint.m_c[1];
			blk.m_endpoints[9] = ccell_results[p1].m_astc_high_endpoint.m_c[1];
			blk.m_endpoints[10] = ccell_results[p1].m_astc_low_endpoint.m_c[2];
			blk.m_endpoints[11] = ccell_results[p1].m_astc_high_endpoint.m_c[2];

			s0 = g_astc_unquant[endpoint_range][blk.m_endpoints[0 + 6]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[2 + 6]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[4 + 6]].m_unquant;
			s1 = g_astc_unquant[endpoint_range][blk.m_endpoints[1 + 6]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[3 + 6]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[5 + 6]].m_unquant;

			if (s1 < s0)
			{
				std::swap(blk.m_endpoints[0 + 6], blk.m_endpoints[1 + 6]);
				std::swap(blk.m_endpoints[2 + 6], blk.m_endpoints[3 + 6]);
				std::swap(blk.m_endpoints[4 + 6], blk.m_endpoints[5 + 6]);
				invert[1] = true;
			}

			for (uint32_t y = 0; y < 4; y++)
			{
				for (uint32_t x = 0; x < 4; x++)
				{
					const uint32_t astc_part = bc7_convert_partition_index_3_to_2(g_bc7_partition3[16 * bc7_pattern + x + y * 4], common_pattern_k);

					blk.m_weights[x + y * 4] = ccell_result_selectors[astc_part][part_pixel_index[y][x]];

					if (invert[astc_part])
						blk.m_weights[x + y * 4] = 3 - blk.m_weights[x + y * 4];
				}
			}

			assert(total_results < MAX_ENCODE_RESULTS);
			if (total_results < MAX_ENCODE_RESULTS)
			{
				pResults[total_results].m_uastc_mode = 7;
				pResults[total_results].m_common_pattern = common_pattern;
				pResults[total_results].m_astc = blk;
				pResults[total_results].m_astc_err = total_part_err;
				total_results++;
			}

		} // common_pattern
	}

	static void estimate_partition2_list(uint32_t num_weights, uint32_t num_comps, const uint32_t* pWeights, const color_rgba block[4][4], uint32_t* pParts, uint32_t max_parts, const uint32_t weights[4])
	{
		assert(pWeights[0] == 0 && pWeights[num_weights - 1] == 64);

		const uint32_t MAX_PARTS = 8;
		assert(max_parts <= MAX_PARTS);

		uint64_t part_error[MAX_PARTS];
		memset(part_error, 0xFF, sizeof(part_error));
		memset(pParts, 0, sizeof(pParts[0]) * max_parts);

		for (uint32_t common_pattern = 0; common_pattern < TOTAL_ASTC_BC7_COMMON_PARTITIONS2; common_pattern++)
		{
			const uint32_t bc7_pattern = g_astc_bc7_common_partitions2[common_pattern].m_bc7;

			const uint8_t* pPartition = &g_bc7_partition2[bc7_pattern * 16];

			color_quad_u8 subset_colors[2][16];
			uint32_t subset_total_colors[2] = { 0, 0 };
			for (uint32_t index = 0; index < 16; index++)
				subset_colors[pPartition[index]][subset_total_colors[pPartition[index]]++] = ((const color_quad_u8*)block)[index];

			uint64_t total_subset_err = 0;
			for (uint32_t subset = 0; subset < 2; subset++)
				total_subset_err += color_cell_compression_est_astc(num_weights, num_comps, pWeights, subset_total_colors[subset], &subset_colors[subset][0], UINT64_MAX, weights);

			for (int i = 0; i < (int)max_parts; i++)
			{
				if (total_subset_err < part_error[i])
				{
					for (int j = max_parts - 1; j > i; --j)
					{
						pParts[j] = pParts[j - 1];
						part_error[j] = part_error[j - 1];
					}

					pParts[i] = common_pattern;
					part_error[i] = total_subset_err;

					break;
				}
			}
		}

#ifdef _DEBUG
		for (uint32_t i = 0; i < max_parts - 1; i++)
		{
			assert(part_error[i] <= part_error[i + 1]);
		}
#endif
	}
		
	// 9. DualPlane: 0, WeightRange: 2 (4), Subsets: 2, CEM: 12 (RGBA Direct), EndpointRange: 8 (16) - BC7 MODE 7
	// 16. DualPlane: 0, WeightRange : 2 (4), Subsets : 2, CEM: 4 (LA Direct), EndpointRange : 20 (256) - BC7 MODE 7
	static void astc_mode9_or_16(uint32_t mode, const color_rgba source_block[4][4], uastc_encode_results* pResults, uint32_t& total_results, bc7enc_compress_block_params& comp_params, uint32_t estimate_partition_list_size)
	{
		assert(mode == 9 || mode == 16);

		const color_rgba* pBlock = &source_block[0][0];

		color_rgba temp_block[16];
		if (mode == 16)
		{
			for (uint32_t i = 0; i < 16; i++)
			{
				if (mode == 16)
				{
					assert(pBlock[i].r == pBlock[i].g);
					assert(pBlock[i].r == pBlock[i].b);
				}

				const uint32_t l = pBlock[i].r;
				const uint32_t a = pBlock[i].a;

				// Use (l,0,0,a) not (l,l,l,a) so both components are treated equally.
				temp_block[i].set_noclamp_rgba(l, 0, 0, a);
			}

			pBlock = temp_block;
		}

		const uint32_t weights[4] = { 1, 1, 1, 1 };

		//const uint32_t weight_range = 2;
		const uint32_t endpoint_range = (mode == 16) ? 20 : 8;

		uint32_t first_common_pattern = 0;
		uint32_t last_common_pattern = TOTAL_ASTC_BC7_COMMON_PARTITIONS2;
		bool use_part_list = false;

		const uint32_t MAX_PARTS = 8;
		uint32_t parts[MAX_PARTS];

		if (estimate_partition_list_size == 1)
		{
			first_common_pattern = estimate_partition2(4, 4, g_bc7_weights2, (const color_rgba(*)[4])pBlock, weights);
			last_common_pattern = first_common_pattern + 1;
		}
		else if (estimate_partition_list_size > 0)
		{
			assert(estimate_partition_list_size <= MAX_PARTS);
			estimate_partition_list_size = basisu::minimum(estimate_partition_list_size, MAX_PARTS);

			estimate_partition2_list(4, 4, g_bc7_weights2, (const color_rgba(*)[4])pBlock, parts, estimate_partition_list_size, weights);

			first_common_pattern = 0;
			last_common_pattern = estimate_partition_list_size;
			use_part_list = true;

#ifdef _DEBUG
			assert(parts[0] == estimate_partition2(4, 4, g_bc7_weights2, (const color_rgba(*)[4])pBlock, weights));
#endif
		}

		for (uint32_t common_pattern_iter = first_common_pattern; common_pattern_iter < last_common_pattern; common_pattern_iter++)
		{
			const uint32_t common_pattern = use_part_list ? parts[common_pattern_iter] : common_pattern_iter;

			const uint32_t bc7_pattern = g_astc_bc7_common_partitions2[common_pattern].m_bc7;

			color_rgba part_pixels[2][16];
			uint32_t part_pixel_index[4][4];
			uint32_t num_part_pixels[2] = { 0, 0 };

			for (uint32_t y = 0; y < 4; y++)
			{
				for (uint32_t x = 0; x < 4; x++)
				{
					const uint32_t part = g_bc7_partition2[16 * bc7_pattern + x + y * 4];
					part_pixel_index[y][x] = num_part_pixels[part];
					part_pixels[part][num_part_pixels[part]++] = pBlock[y * 4 + x];
				}
			}

			color_cell_compressor_params ccell_params[2];
			color_cell_compressor_results ccell_results[2];
			uint8_t ccell_result_selectors[2][16];
			uint8_t ccell_result_selectors_temp[2][16];

			uint64_t total_err = 0;
			for (uint32_t subset = 0; subset < 2; subset++)
			{
				memset(&ccell_params[subset], 0, sizeof(ccell_params[subset]));

				ccell_params[subset].m_num_pixels = num_part_pixels[subset];
				ccell_params[subset].m_pPixels = (color_quad_u8*)&part_pixels[subset][0];
				ccell_params[subset].m_num_selector_weights = 4;
				ccell_params[subset].m_pSelector_weights = g_bc7_weights2;
				ccell_params[subset].m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights2x;
				ccell_params[subset].m_astc_endpoint_range = endpoint_range;
				ccell_params[subset].m_weights[0] = weights[0];
				ccell_params[subset].m_weights[1] = weights[1];
				ccell_params[subset].m_weights[2] = weights[2];
				ccell_params[subset].m_weights[3] = weights[3];
				ccell_params[subset].m_has_alpha = true;

				memset(&ccell_results[subset], 0, sizeof(ccell_results[subset]));
				ccell_results[subset].m_pSelectors = &ccell_result_selectors[subset][0];
				ccell_results[subset].m_pSelectors_temp = &ccell_result_selectors_temp[subset][0];

				uint64_t subset_err = color_cell_compression(255, &ccell_params[subset], &ccell_results[subset], &comp_params);

				if (mode == 16)
				{
					color_rgba colors[4];
					for (uint32_t c = 0; c < 4; c++)
					{
						colors[0].m_comps[c] = g_astc_unquant[endpoint_range][ccell_results[subset].m_astc_low_endpoint.m_c[(c < 3) ? 0 : 3]].m_unquant;
						colors[3].m_comps[c] = g_astc_unquant[endpoint_range][ccell_results[subset].m_astc_high_endpoint.m_c[(c < 3) ? 0 : 3]].m_unquant;
					}

					for (uint32_t i = 1; i < 4 - 1; i++)
						for (uint32_t c = 0; c < 4; c++)
							colors[i].m_comps[c] = (uint8_t)astc_interpolate(colors[0].m_comps[c], colors[3].m_comps[c], g_bc7_weights2[i], false);

					for (uint32_t p = 0; p < ccell_params[subset].m_num_pixels; p++)
					{
						color_rgba orig_pix(part_pixels[subset][p]);
						orig_pix.g = orig_pix.r;
						orig_pix.b = orig_pix.r;
						total_err += color_distance_la(orig_pix, colors[ccell_result_selectors[subset][p]]);
					}
				}
				else
				{
					total_err += subset_err;
				}
			} // subset

			// ASTC
			astc_block_desc astc_results;
			memset(&astc_results, 0, sizeof(astc_results));

			astc_results.m_dual_plane = false;
			astc_results.m_weight_range = 2;

			astc_results.m_ccs = 0;
			astc_results.m_subsets = 2;
			astc_results.m_partition_seed = g_astc_bc7_common_partitions2[common_pattern].m_astc;
			astc_results.m_cem = (mode == 16) ? 4 : 12;

			uint32_t part[2] = { 0, 1 };
			if (g_astc_bc7_common_partitions2[common_pattern].m_invert)
				std::swap(part[0], part[1]);

			bool invert[2] = { false, false };

			for (uint32_t p = 0; p < 2; p++)
			{
				if (mode == 16)
				{
					astc_results.m_endpoints[p * 4 + 0] = ccell_results[part[p]].m_astc_low_endpoint.m_c[0];
					astc_results.m_endpoints[p * 4 + 1] = ccell_results[part[p]].m_astc_high_endpoint.m_c[0];

					astc_results.m_endpoints[p * 4 + 2] = ccell_results[part[p]].m_astc_low_endpoint.m_c[3];
					astc_results.m_endpoints[p * 4 + 3] = ccell_results[part[p]].m_astc_high_endpoint.m_c[3];
				}
				else
				{
					for (uint32_t c = 0; c < 4; c++)
					{
						astc_results.m_endpoints[p * 8 + c * 2] = ccell_results[part[p]].m_astc_low_endpoint.m_c[c];
						astc_results.m_endpoints[p * 8 + c * 2 + 1] = ccell_results[part[p]].m_astc_high_endpoint.m_c[c];
					}

					int s0 = g_astc_unquant[endpoint_range][astc_results.m_endpoints[p * 8 + 0]].m_unquant +
						g_astc_unquant[endpoint_range][astc_results.m_endpoints[p * 8 + 2]].m_unquant +
						g_astc_unquant[endpoint_range][astc_results.m_endpoints[p * 8 + 4]].m_unquant;

					int s1 = g_astc_unquant[endpoint_range][astc_results.m_endpoints[p * 8 + 1]].m_unquant +
						g_astc_unquant[endpoint_range][astc_results.m_endpoints[p * 8 + 3]].m_unquant +
						g_astc_unquant[endpoint_range][astc_results.m_endpoints[p * 8 + 5]].m_unquant;

					if (s1 < s0)
					{
						std::swap(astc_results.m_endpoints[p * 8 + 0], astc_results.m_endpoints[p * 8 + 1]);
						std::swap(astc_results.m_endpoints[p * 8 + 2], astc_results.m_endpoints[p * 8 + 3]);
						std::swap(astc_results.m_endpoints[p * 8 + 4], astc_results.m_endpoints[p * 8 + 5]);
						std::swap(astc_results.m_endpoints[p * 8 + 6], astc_results.m_endpoints[p * 8 + 7]);
						invert[p] = true;
					}
				}
			}

			for (uint32_t y = 0; y < 4; y++)
			{
				for (uint32_t x = 0; x < 4; x++)
				{
					const uint32_t bc7_part = g_bc7_partition2[16 * bc7_pattern + x + y * 4];

					astc_results.m_weights[x + y * 4] = ccell_result_selectors[bc7_part][part_pixel_index[y][x]];

					uint32_t astc_part = bc7_part;
					if (g_astc_bc7_common_partitions2[common_pattern].m_invert)
						astc_part = 1 - astc_part;

					if (invert[astc_part])
						astc_results.m_weights[x + y * 4] = 3 - astc_results.m_weights[x + y * 4];
				}
			}

			assert(total_results < MAX_ENCODE_RESULTS);
			if (total_results < MAX_ENCODE_RESULTS)
			{
				pResults[total_results].m_uastc_mode = mode;
				pResults[total_results].m_common_pattern = common_pattern;
				pResults[total_results].m_astc = astc_results;
				pResults[total_results].m_astc_err = total_err;
				total_results++;
			}

		} // common_pattern
	}

	// MODE 10
	// DualPlane: 0, WeightRange: 8 (16), Subsets: 1, CEM: 12 (RGBA Direct      ), EndpointRange: 13 (48)       MODE6
	static void astc_mode10(const color_rgba block[4][4], uastc_encode_results* pResults, uint32_t& total_results, bc7enc_compress_block_params& comp_params)
	{
		const uint32_t weight_range = 8;
		const uint32_t endpoint_range = 13;

		color_cell_compressor_params ccell_params;
		memset(&ccell_params, 0, sizeof(ccell_params));

		ccell_params.m_num_pixels = 16;
		ccell_params.m_pPixels = (color_quad_u8*)&block[0][0];
		ccell_params.m_num_selector_weights = 16;
		ccell_params.m_pSelector_weights = g_astc_weights4;
		ccell_params.m_pSelector_weightsx = (const bc7enc_vec4F*)g_astc_weights4x;
		ccell_params.m_astc_endpoint_range = endpoint_range;
		ccell_params.m_weights[0] = 1;
		ccell_params.m_weights[1] = 1;
		ccell_params.m_weights[2] = 1;
		ccell_params.m_weights[3] = 1;
		ccell_params.m_has_alpha = true;

		color_cell_compressor_results ccell_results;
		uint8_t ccell_result_selectors[16];
		uint8_t ccell_result_selectors_temp[16];
		memset(&ccell_results, 0, sizeof(ccell_results));
		ccell_results.m_pSelectors = &ccell_result_selectors[0];
		ccell_results.m_pSelectors_temp = &ccell_result_selectors_temp[0];

		uint64_t part_err = color_cell_compression(255, &ccell_params, &ccell_results, &comp_params);

		// ASTC
		astc_block_desc astc_results;
		memset(&astc_results, 0, sizeof(astc_results));

		astc_results.m_dual_plane = false;
		astc_results.m_weight_range = weight_range;

		astc_results.m_ccs = 0;
		astc_results.m_subsets = 1;
		astc_results.m_partition_seed = 0;
		astc_results.m_cem = 12;

		astc_results.m_endpoints[0] = ccell_results.m_astc_low_endpoint.m_c[0];
		astc_results.m_endpoints[1] = ccell_results.m_astc_high_endpoint.m_c[0];
		astc_results.m_endpoints[2] = ccell_results.m_astc_low_endpoint.m_c[1];
		astc_results.m_endpoints[3] = ccell_results.m_astc_high_endpoint.m_c[1];
		astc_results.m_endpoints[4] = ccell_results.m_astc_low_endpoint.m_c[2];
		astc_results.m_endpoints[5] = ccell_results.m_astc_high_endpoint.m_c[2];
		astc_results.m_endpoints[6] = ccell_results.m_astc_low_endpoint.m_c[3];
		astc_results.m_endpoints[7] = ccell_results.m_astc_high_endpoint.m_c[3];

		bool invert = false;

		int s0 = g_astc_unquant[endpoint_range][astc_results.m_endpoints[0]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[2]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[4]].m_unquant;
		int s1 = g_astc_unquant[endpoint_range][astc_results.m_endpoints[1]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[3]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[5]].m_unquant;
		if (s1 < s0)
		{
			std::swap(astc_results.m_endpoints[0], astc_results.m_endpoints[1]);
			std::swap(astc_results.m_endpoints[2], astc_results.m_endpoints[3]);
			std::swap(astc_results.m_endpoints[4], astc_results.m_endpoints[5]);
			std::swap(astc_results.m_endpoints[6], astc_results.m_endpoints[7]);
			invert = true;
		}

		for (uint32_t y = 0; y < 4; y++)
		{
			for (uint32_t x = 0; x < 4; x++)
			{
				astc_results.m_weights[x + y * 4] = ccell_result_selectors[x + y * 4];

				if (invert)
					astc_results.m_weights[x + y * 4] = 15 - astc_results.m_weights[x + y * 4];
			}
		}

		assert(total_results < MAX_ENCODE_RESULTS);
		if (total_results < MAX_ENCODE_RESULTS)
		{
			pResults[total_results].m_uastc_mode = 10;
			pResults[total_results].m_common_pattern = 0;
			pResults[total_results].m_astc = astc_results;
			pResults[total_results].m_astc_err = part_err;
			total_results++;
		}
	}

	// 11. DualPlane: 1, WeightRange: 2 (4), Subsets: 1, CEM: 12 (RGBA Direct), EndpointRange: 13 (48)        MODE5
	// 17. DualPlane: 1, WeightRange : 2 (4), Subsets : 1, CEM : 4 (LA Direct), EndpointRange : 20 (256)    BC7 MODE5
	static void astc_mode11_or_17(uint32_t mode, const color_rgba block[4][4], uastc_encode_results* pResults, uint32_t& total_results, bc7enc_compress_block_params& comp_params)
	{
		assert((mode == 11) || (mode == 17));

		const uint32_t weight_range = 2;
		const uint32_t endpoint_range = (mode == 17) ? 20 : 13;

		bc7enc_compress_block_params local_comp_params(comp_params);
		local_comp_params.m_perceptual = false;
		local_comp_params.m_weights[0] = 1;
		local_comp_params.m_weights[1] = 1;
		local_comp_params.m_weights[2] = 1;
		local_comp_params.m_weights[3] = 1;

		const uint32_t last_rot_comp = (mode == 17) ? 1 : 4;

		for (uint32_t rot_comp = 0; rot_comp < last_rot_comp; rot_comp++)
		{
			color_quad_u8 block_rgb[16];
			color_quad_u8 block_a[16];
			for (uint32_t i = 0; i < 16; i++)
			{
				block_rgb[i] = ((color_quad_u8*)&block[0][0])[i];
				block_a[i] = block_rgb[i];

				if (mode == 17)
				{
					assert(block_rgb[i].m_c[0] == block_rgb[i].m_c[1]);
					assert(block_rgb[i].m_c[0] == block_rgb[i].m_c[2]);

					block_a[i].m_c[0] = block_rgb[i].m_c[3];
					block_a[i].m_c[1] = block_rgb[i].m_c[3];
					block_a[i].m_c[2] = block_rgb[i].m_c[3];
					block_a[i].m_c[3] = 255;

					block_rgb[i].m_c[1] = block_rgb[i].m_c[0];
					block_rgb[i].m_c[2] = block_rgb[i].m_c[0];
					block_rgb[i].m_c[3] = 255;
				}
				else
				{
					uint8_t c = block_a[i].m_c[rot_comp];
					block_a[i].m_c[0] = c;
					block_a[i].m_c[1] = c;
					block_a[i].m_c[2] = c;
					block_a[i].m_c[3] = 255;

					block_rgb[i].m_c[rot_comp] = block_rgb[i].m_c[3];
					block_rgb[i].m_c[3] = 255;
				}
			}

			uint8_t ccell_result_selectors_temp[16];

			color_cell_compressor_params ccell_params_rgb;
			memset(&ccell_params_rgb, 0, sizeof(ccell_params_rgb));

			ccell_params_rgb.m_num_pixels = 16;
			ccell_params_rgb.m_pPixels = block_rgb;
			ccell_params_rgb.m_num_selector_weights = 4;
			ccell_params_rgb.m_pSelector_weights = g_bc7_weights2;
			ccell_params_rgb.m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights2x;
			ccell_params_rgb.m_astc_endpoint_range = endpoint_range;
			ccell_params_rgb.m_weights[0] = 1;
			ccell_params_rgb.m_weights[1] = 1;
			ccell_params_rgb.m_weights[2] = 1;
			ccell_params_rgb.m_weights[3] = 1;

			color_cell_compressor_results ccell_results_rgb;
			uint8_t ccell_result_selectors_rgb[16];
			memset(&ccell_results_rgb, 0, sizeof(ccell_results_rgb));
			ccell_results_rgb.m_pSelectors = &ccell_result_selectors_rgb[0];
			ccell_results_rgb.m_pSelectors_temp = &ccell_result_selectors_temp[0];

			uint64_t part_err_rgb = color_cell_compression(255, &ccell_params_rgb, &ccell_results_rgb, &local_comp_params);

			color_cell_compressor_params ccell_params_a;
			memset(&ccell_params_a, 0, sizeof(ccell_params_a));

			ccell_params_a.m_num_pixels = 16;
			ccell_params_a.m_pPixels = block_a;
			ccell_params_a.m_num_selector_weights = 4;
			ccell_params_a.m_pSelector_weights = g_bc7_weights2;
			ccell_params_a.m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights2x;
			ccell_params_a.m_astc_endpoint_range = endpoint_range;
			ccell_params_a.m_weights[0] = 1;
			ccell_params_a.m_weights[1] = 1;
			ccell_params_a.m_weights[2] = 1;
			ccell_params_a.m_weights[3] = 1;

			color_cell_compressor_results ccell_results_a;
			uint8_t ccell_result_selectors_a[16];
			memset(&ccell_results_a, 0, sizeof(ccell_results_a));
			ccell_results_a.m_pSelectors = &ccell_result_selectors_a[0];
			ccell_results_a.m_pSelectors_temp = &ccell_result_selectors_temp[0];

			uint64_t part_err_a = color_cell_compression(255, &ccell_params_a, &ccell_results_a, &local_comp_params) / 3;

			uint64_t total_err = (mode == 17) ? ((part_err_rgb / 3) + part_err_a) : (part_err_rgb + part_err_a);

			// ASTC
			astc_block_desc blk;
			memset(&blk, 0, sizeof(blk));

			blk.m_dual_plane = true;
			blk.m_weight_range = weight_range;

			blk.m_ccs = (mode == 17) ? 3 : rot_comp;
			blk.m_subsets = 1;
			blk.m_partition_seed = 0;
			blk.m_cem = (mode == 17) ? 4 : 12;

			bool invert = false;

			if (mode == 17)
			{
				assert(ccell_results_rgb.m_astc_low_endpoint.m_c[0] == ccell_results_rgb.m_astc_low_endpoint.m_c[1]);
				assert(ccell_results_rgb.m_astc_low_endpoint.m_c[0] == ccell_results_rgb.m_astc_low_endpoint.m_c[2]);

				assert(ccell_results_rgb.m_astc_high_endpoint.m_c[0] == ccell_results_rgb.m_astc_high_endpoint.m_c[1]);
				assert(ccell_results_rgb.m_astc_high_endpoint.m_c[0] == ccell_results_rgb.m_astc_high_endpoint.m_c[2]);

				blk.m_endpoints[0] = ccell_results_rgb.m_astc_low_endpoint.m_c[0];
				blk.m_endpoints[1] = ccell_results_rgb.m_astc_high_endpoint.m_c[0];

				blk.m_endpoints[2] = ccell_results_a.m_astc_low_endpoint.m_c[0];
				blk.m_endpoints[3] = ccell_results_a.m_astc_high_endpoint.m_c[0];
			}
			else
			{
				blk.m_endpoints[0] = (rot_comp == 0 ? ccell_results_a : ccell_results_rgb).m_astc_low_endpoint.m_c[0];
				blk.m_endpoints[1] = (rot_comp == 0 ? ccell_results_a : ccell_results_rgb).m_astc_high_endpoint.m_c[0];
				blk.m_endpoints[2] = (rot_comp == 1 ? ccell_results_a : ccell_results_rgb).m_astc_low_endpoint.m_c[1];
				blk.m_endpoints[3] = (rot_comp == 1 ? ccell_results_a : ccell_results_rgb).m_astc_high_endpoint.m_c[1];
				blk.m_endpoints[4] = (rot_comp == 2 ? ccell_results_a : ccell_results_rgb).m_astc_low_endpoint.m_c[2];
				blk.m_endpoints[5] = (rot_comp == 2 ? ccell_results_a : ccell_results_rgb).m_astc_high_endpoint.m_c[2];
				if (rot_comp == 3)
				{
					blk.m_endpoints[6] = ccell_results_a.m_astc_low_endpoint.m_c[0];
					blk.m_endpoints[7] = ccell_results_a.m_astc_high_endpoint.m_c[0];
				}
				else
				{
					blk.m_endpoints[6] = ccell_results_rgb.m_astc_low_endpoint.m_c[rot_comp];
					blk.m_endpoints[7] = ccell_results_rgb.m_astc_high_endpoint.m_c[rot_comp];
				}

				int s0 = g_astc_unquant[endpoint_range][blk.m_endpoints[0]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[2]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[4]].m_unquant;
				int s1 = g_astc_unquant[endpoint_range][blk.m_endpoints[1]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[3]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[5]].m_unquant;
				if (s1 < s0)
				{
					std::swap(blk.m_endpoints[0], blk.m_endpoints[1]);
					std::swap(blk.m_endpoints[2], blk.m_endpoints[3]);
					std::swap(blk.m_endpoints[4], blk.m_endpoints[5]);
					std::swap(blk.m_endpoints[6], blk.m_endpoints[7]);
					invert = true;
				}
			}

			for (uint32_t y = 0; y < 4; y++)
			{
				for (uint32_t x = 0; x < 4; x++)
				{
					uint32_t rgb_index = ccell_result_selectors_rgb[x + y * 4];
					uint32_t a_index = ccell_result_selectors_a[x + y * 4];

					if (invert)
					{
						rgb_index = 3 - rgb_index;
						a_index = 3 - a_index;
					}

					blk.m_weights[(x + y * 4) * 2 + 0] = (uint8_t)rgb_index;
					blk.m_weights[(x + y * 4) * 2 + 1] = (uint8_t)a_index;
				}
			}

			assert(total_results < MAX_ENCODE_RESULTS);
			if (total_results < MAX_ENCODE_RESULTS)
			{
				pResults[total_results].m_uastc_mode = mode;
				pResults[total_results].m_common_pattern = 0;
				pResults[total_results].m_astc = blk;
				pResults[total_results].m_astc_err = total_err;
				total_results++;
			}
		} // rot_comp
	}

	// MODE 12
	// DualPlane: 0, WeightRange: 5 (8), Subsets: 1, CEM: 12 (RGBA Direct      ), EndpointRange: 19 (192)       MODE6
	static void astc_mode12(const color_rgba block[4][4], uastc_encode_results* pResults, uint32_t& total_results, bc7enc_compress_block_params& comp_params)
	{
		const uint32_t weight_range = 5;
		const uint32_t endpoint_range = 19;

		color_cell_compressor_params ccell_params;
		memset(&ccell_params, 0, sizeof(ccell_params));

		ccell_params.m_num_pixels = 16;
		ccell_params.m_pPixels = (color_quad_u8*)&block[0][0];
		ccell_params.m_num_selector_weights = 8;
		ccell_params.m_pSelector_weights = g_bc7_weights3;
		ccell_params.m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights3x;
		ccell_params.m_astc_endpoint_range = endpoint_range;
		ccell_params.m_weights[0] = 1;
		ccell_params.m_weights[1] = 1;
		ccell_params.m_weights[2] = 1;
		ccell_params.m_weights[3] = 1;
		ccell_params.m_has_alpha = true;

		color_cell_compressor_results ccell_results;
		uint8_t ccell_result_selectors[16];
		uint8_t ccell_result_selectors_temp[16];
		memset(&ccell_results, 0, sizeof(ccell_results));
		ccell_results.m_pSelectors = &ccell_result_selectors[0];
		ccell_results.m_pSelectors_temp = &ccell_result_selectors_temp[0];

		uint64_t part_err = color_cell_compression(255, &ccell_params, &ccell_results, &comp_params);

		// ASTC
		astc_block_desc astc_results;
		memset(&astc_results, 0, sizeof(astc_results));

		astc_results.m_dual_plane = false;
		astc_results.m_weight_range = weight_range;

		astc_results.m_ccs = 0;
		astc_results.m_subsets = 1;
		astc_results.m_partition_seed = 0;
		astc_results.m_cem = 12;

		astc_results.m_endpoints[0] = ccell_results.m_astc_low_endpoint.m_c[0];
		astc_results.m_endpoints[1] = ccell_results.m_astc_high_endpoint.m_c[0];
		astc_results.m_endpoints[2] = ccell_results.m_astc_low_endpoint.m_c[1];
		astc_results.m_endpoints[3] = ccell_results.m_astc_high_endpoint.m_c[1];
		astc_results.m_endpoints[4] = ccell_results.m_astc_low_endpoint.m_c[2];
		astc_results.m_endpoints[5] = ccell_results.m_astc_high_endpoint.m_c[2];
		astc_results.m_endpoints[6] = ccell_results.m_astc_low_endpoint.m_c[3];
		astc_results.m_endpoints[7] = ccell_results.m_astc_high_endpoint.m_c[3];

		bool invert = false;

		int s0 = g_astc_unquant[endpoint_range][astc_results.m_endpoints[0]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[2]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[4]].m_unquant;
		int s1 = g_astc_unquant[endpoint_range][astc_results.m_endpoints[1]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[3]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[5]].m_unquant;
		if (s1 < s0)
		{
			std::swap(astc_results.m_endpoints[0], astc_results.m_endpoints[1]);
			std::swap(astc_results.m_endpoints[2], astc_results.m_endpoints[3]);
			std::swap(astc_results.m_endpoints[4], astc_results.m_endpoints[5]);
			std::swap(astc_results.m_endpoints[6], astc_results.m_endpoints[7]);
			invert = true;
		}

		for (uint32_t y = 0; y < 4; y++)
		{
			for (uint32_t x = 0; x < 4; x++)
			{
				astc_results.m_weights[x + y * 4] = ccell_result_selectors[x + y * 4];

				if (invert)
					astc_results.m_weights[x + y * 4] = 7 - astc_results.m_weights[x + y * 4];
			}
		}

		assert(total_results < MAX_ENCODE_RESULTS);
		if (total_results < MAX_ENCODE_RESULTS)
		{
			pResults[total_results].m_uastc_mode = 12;
			pResults[total_results].m_common_pattern = 0;
			pResults[total_results].m_astc = astc_results;
			pResults[total_results].m_astc_err = part_err;
			total_results++;
		}
	}

	// 13. DualPlane: 1, WeightRange: 0 (2), Subsets: 1, CEM: 12 (RGBA Direct      ), EndpointRange: 20 (256)        MODE5
	static void astc_mode13(const color_rgba block[4][4], uastc_encode_results* pResults, uint32_t& total_results, bc7enc_compress_block_params& comp_params)
	{
		bc7enc_compress_block_params local_comp_params(comp_params);
		local_comp_params.m_perceptual = false;
		local_comp_params.m_weights[0] = 1;
		local_comp_params.m_weights[1] = 1;
		local_comp_params.m_weights[2] = 1;
		local_comp_params.m_weights[3] = 1;

		for (uint32_t rot_comp = 0; rot_comp < 4; rot_comp++)
		{
			const uint32_t weight_range = 0;
			const uint32_t endpoint_range = 20;

			color_quad_u8 block_rgb[16];
			color_quad_u8 block_a[16];
			for (uint32_t i = 0; i < 16; i++)
			{
				block_rgb[i] = ((color_quad_u8*)&block[0][0])[i];
				block_a[i] = block_rgb[i];

				uint8_t c = block_a[i].m_c[rot_comp];
				block_a[i].m_c[0] = c;
				block_a[i].m_c[1] = c;
				block_a[i].m_c[2] = c;
				block_a[i].m_c[3] = 255;

				block_rgb[i].m_c[rot_comp] = block_rgb[i].m_c[3];
				block_rgb[i].m_c[3] = 255;
			}

			uint8_t ccell_result_selectors_temp[16];

			color_cell_compressor_params ccell_params_rgb;
			memset(&ccell_params_rgb, 0, sizeof(ccell_params_rgb));

			ccell_params_rgb.m_num_pixels = 16;
			ccell_params_rgb.m_pPixels = block_rgb;
			ccell_params_rgb.m_num_selector_weights = 2;
			ccell_params_rgb.m_pSelector_weights = g_bc7_weights1;
			ccell_params_rgb.m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights1x;
			ccell_params_rgb.m_astc_endpoint_range = endpoint_range;
			ccell_params_rgb.m_weights[0] = 1;
			ccell_params_rgb.m_weights[1] = 1;
			ccell_params_rgb.m_weights[2] = 1;
			ccell_params_rgb.m_weights[3] = 1;

			color_cell_compressor_results ccell_results_rgb;
			uint8_t ccell_result_selectors_rgb[16];
			memset(&ccell_results_rgb, 0, sizeof(ccell_results_rgb));
			ccell_results_rgb.m_pSelectors = &ccell_result_selectors_rgb[0];
			ccell_results_rgb.m_pSelectors_temp = &ccell_result_selectors_temp[0];

			uint64_t part_err_rgb = color_cell_compression(255, &ccell_params_rgb, &ccell_results_rgb, &local_comp_params);

			color_cell_compressor_params ccell_params_a;
			memset(&ccell_params_a, 0, sizeof(ccell_params_a));

			ccell_params_a.m_num_pixels = 16;
			ccell_params_a.m_pPixels = block_a;
			ccell_params_a.m_num_selector_weights = 2;
			ccell_params_a.m_pSelector_weights = g_bc7_weights1;
			ccell_params_a.m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights1x;
			ccell_params_a.m_astc_endpoint_range = endpoint_range;
			ccell_params_a.m_weights[0] = 1;
			ccell_params_a.m_weights[1] = 1;
			ccell_params_a.m_weights[2] = 1;
			ccell_params_a.m_weights[3] = 1;

			color_cell_compressor_results ccell_results_a;
			uint8_t ccell_result_selectors_a[16];
			memset(&ccell_results_a, 0, sizeof(ccell_results_a));
			ccell_results_a.m_pSelectors = &ccell_result_selectors_a[0];
			ccell_results_a.m_pSelectors_temp = &ccell_result_selectors_temp[0];

			uint64_t part_err_a = color_cell_compression(255, &ccell_params_a, &ccell_results_a, &local_comp_params) / 3;

			uint64_t total_err = part_err_rgb + part_err_a;

			// ASTC
			astc_block_desc blk;
			memset(&blk, 0, sizeof(blk));

			blk.m_dual_plane = true;
			blk.m_weight_range = weight_range;

			blk.m_ccs = rot_comp;
			blk.m_subsets = 1;
			blk.m_partition_seed = 0;
			blk.m_cem = 12;

			blk.m_endpoints[0] = (rot_comp == 0 ? ccell_results_a : ccell_results_rgb).m_astc_low_endpoint.m_c[0];
			blk.m_endpoints[1] = (rot_comp == 0 ? ccell_results_a : ccell_results_rgb).m_astc_high_endpoint.m_c[0];
			blk.m_endpoints[2] = (rot_comp == 1 ? ccell_results_a : ccell_results_rgb).m_astc_low_endpoint.m_c[1];
			blk.m_endpoints[3] = (rot_comp == 1 ? ccell_results_a : ccell_results_rgb).m_astc_high_endpoint.m_c[1];
			blk.m_endpoints[4] = (rot_comp == 2 ? ccell_results_a : ccell_results_rgb).m_astc_low_endpoint.m_c[2];
			blk.m_endpoints[5] = (rot_comp == 2 ? ccell_results_a : ccell_results_rgb).m_astc_high_endpoint.m_c[2];
			if (rot_comp == 3)
			{
				blk.m_endpoints[6] = ccell_results_a.m_astc_low_endpoint.m_c[0];
				blk.m_endpoints[7] = ccell_results_a.m_astc_high_endpoint.m_c[0];
			}
			else
			{
				blk.m_endpoints[6] = ccell_results_rgb.m_astc_low_endpoint.m_c[rot_comp];
				blk.m_endpoints[7] = ccell_results_rgb.m_astc_high_endpoint.m_c[rot_comp];
			}

			bool invert = false;

			int s0 = g_astc_unquant[endpoint_range][blk.m_endpoints[0]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[2]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[4]].m_unquant;
			int s1 = g_astc_unquant[endpoint_range][blk.m_endpoints[1]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[3]].m_unquant + g_astc_unquant[endpoint_range][blk.m_endpoints[5]].m_unquant;
			if (s1 < s0)
			{
				std::swap(blk.m_endpoints[0], blk.m_endpoints[1]);
				std::swap(blk.m_endpoints[2], blk.m_endpoints[3]);
				std::swap(blk.m_endpoints[4], blk.m_endpoints[5]);
				std::swap(blk.m_endpoints[6], blk.m_endpoints[7]);
				invert = true;
			}

			for (uint32_t y = 0; y < 4; y++)
			{
				for (uint32_t x = 0; x < 4; x++)
				{
					uint32_t rgb_index = ccell_result_selectors_rgb[x + y * 4];
					uint32_t a_index = ccell_result_selectors_a[x + y * 4];

					if (invert)
					{
						rgb_index = 1 - rgb_index;
						a_index = 1 - a_index;
					}

					blk.m_weights[(x + y * 4) * 2 + 0] = (uint8_t)rgb_index;
					blk.m_weights[(x + y * 4) * 2 + 1] = (uint8_t)a_index;
				}
			}

			assert(total_results < MAX_ENCODE_RESULTS);
			if (total_results < MAX_ENCODE_RESULTS)
			{
				pResults[total_results].m_uastc_mode = 13;
				pResults[total_results].m_common_pattern = 0;
				pResults[total_results].m_astc = blk;
				pResults[total_results].m_astc_err = total_err;
				total_results++;
			}
		} // rot_comp
	}

	// MODE14
	// DualPlane: 0, WeightRange: 2 (4), Subsets: 1, CEM: 12 (RGBA Direct      ), EndpointRange: 20 (256)		MODE6
	static void astc_mode14(const color_rgba block[4][4], uastc_encode_results* pResults, uint32_t& total_results, bc7enc_compress_block_params& comp_params)
	{
		const uint32_t weight_range = 2;
		const uint32_t endpoint_range = 20;

		color_cell_compressor_params ccell_params;
		memset(&ccell_params, 0, sizeof(ccell_params));

		ccell_params.m_num_pixels = 16;
		ccell_params.m_pPixels = (color_quad_u8*)&block[0][0];
		ccell_params.m_num_selector_weights = 4;
		ccell_params.m_pSelector_weights = g_bc7_weights2;
		ccell_params.m_pSelector_weightsx = (const bc7enc_vec4F*)g_bc7_weights2x;
		ccell_params.m_astc_endpoint_range = endpoint_range;
		ccell_params.m_weights[0] = 1;
		ccell_params.m_weights[1] = 1;
		ccell_params.m_weights[2] = 1;
		ccell_params.m_weights[3] = 1;
		ccell_params.m_has_alpha = true;

		color_cell_compressor_results ccell_results;
		uint8_t ccell_result_selectors[16];
		uint8_t ccell_result_selectors_temp[16];
		memset(&ccell_results, 0, sizeof(ccell_results));
		ccell_results.m_pSelectors = &ccell_result_selectors[0];
		ccell_results.m_pSelectors_temp = &ccell_result_selectors_temp[0];

		uint64_t part_err = color_cell_compression(255, &ccell_params, &ccell_results, &comp_params);

		// ASTC
		astc_block_desc astc_results;
		memset(&astc_results, 0, sizeof(astc_results));

		astc_results.m_dual_plane = false;
		astc_results.m_weight_range = weight_range;

		astc_results.m_ccs = 0;
		astc_results.m_subsets = 1;
		astc_results.m_partition_seed = 0;
		astc_results.m_cem = 12;

		astc_results.m_endpoints[0] = ccell_results.m_astc_low_endpoint.m_c[0];
		astc_results.m_endpoints[1] = ccell_results.m_astc_high_endpoint.m_c[0];
		astc_results.m_endpoints[2] = ccell_results.m_astc_low_endpoint.m_c[1];
		astc_results.m_endpoints[3] = ccell_results.m_astc_high_endpoint.m_c[1];
		astc_results.m_endpoints[4] = ccell_results.m_astc_low_endpoint.m_c[2];
		astc_results.m_endpoints[5] = ccell_results.m_astc_high_endpoint.m_c[2];
		astc_results.m_endpoints[6] = ccell_results.m_astc_low_endpoint.m_c[3];
		astc_results.m_endpoints[7] = ccell_results.m_astc_high_endpoint.m_c[3];

		bool invert = false;

		int s0 = g_astc_unquant[endpoint_range][astc_results.m_endpoints[0]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[2]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[4]].m_unquant;
		int s1 = g_astc_unquant[endpoint_range][astc_results.m_endpoints[1]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[3]].m_unquant + g_astc_unquant[endpoint_range][astc_results.m_endpoints[5]].m_unquant;
		if (s1 < s0)
		{
			std::swap(astc_results.m_endpoints[0], astc_results.m_endpoints[1]);
			std::swap(astc_results.m_endpoints[2], astc_results.m_endpoints[3]);
			std::swap(astc_results.m_endpoints[4], astc_results.m_endpoints[5]);
			std::swap(astc_results.m_endpoints[6], astc_results.m_endpoints[7]);
			invert = true;
		}

		for (uint32_t y = 0; y < 4; y++)
		{
			for (uint32_t x = 0; x < 4; x++)
			{
				astc_results.m_weights[x + y * 4] = ccell_result_selectors[x + y * 4];

				if (invert)
					astc_results.m_weights[x + y * 4] = 3 - astc_results.m_weights[x + y * 4];
			}
		}

		assert(total_results < MAX_ENCODE_RESULTS);
		if (total_results < MAX_ENCODE_RESULTS)
		{
			pResults[total_results].m_uastc_mode = 14;
			pResults[total_results].m_common_pattern = 0;
			pResults[total_results