diff options
Diffstat (limited to 'thirdparty/astcenc/astcenc_entry.cpp')
| -rw-r--r-- | thirdparty/astcenc/astcenc_entry.cpp | 109 |
1 files changed, 65 insertions, 44 deletions
diff --git a/thirdparty/astcenc/astcenc_entry.cpp b/thirdparty/astcenc/astcenc_entry.cpp index e53762c26a..71efe9cec4 100644 --- a/thirdparty/astcenc/astcenc_entry.cpp +++ b/thirdparty/astcenc/astcenc_entry.cpp @@ -1,6 +1,6 @@ // SPDX-License-Identifier: Apache-2.0 // ---------------------------------------------------------------------------- -// Copyright 2011-2023 Arm Limited +// Copyright 2011-2024 Arm Limited // // 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 @@ -55,6 +55,7 @@ struct astcenc_preset_config float tune_2partition_early_out_limit_factor; float tune_3partition_early_out_limit_factor; float tune_2plane_early_out_limit_correlation; + float tune_search_mode0_enable; }; /** @@ -63,22 +64,22 @@ struct astcenc_preset_config static const std::array<astcenc_preset_config, 6> preset_configs_high {{ { ASTCENC_PRE_FASTEST, - 2, 10, 6, 4, 43, 2, 2, 2, 2, 2, 85.2f, 63.2f, 3.5f, 1.0f, 1.0f, 0.85f + 2, 10, 6, 4, 43, 2, 2, 2, 2, 2, 85.2f, 63.2f, 3.5f, 1.0f, 1.0f, 0.85f, 0.0f }, { ASTCENC_PRE_FAST, - 3, 18, 10, 8, 55, 3, 3, 2, 2, 2, 85.2f, 63.2f, 3.5f, 1.0f, 1.0f, 0.90f + 3, 18, 10, 8, 55, 3, 3, 2, 2, 2, 85.2f, 63.2f, 3.5f, 1.0f, 1.0f, 0.90f, 0.0f }, { ASTCENC_PRE_MEDIUM, - 4, 34, 28, 16, 77, 3, 3, 2, 2, 2, 95.0f, 70.0f, 2.5f, 1.1f, 1.05f, 0.95f + 4, 34, 28, 16, 77, 3, 3, 2, 2, 2, 95.0f, 70.0f, 2.5f, 1.1f, 1.05f, 0.95f, 0.0f }, { ASTCENC_PRE_THOROUGH, - 4, 82, 60, 30, 94, 4, 4, 3, 2, 2, 105.0f, 77.0f, 10.0f, 1.35f, 1.15f, 0.97f + 4, 82, 60, 30, 94, 4, 4, 3, 2, 2, 105.0f, 77.0f, 10.0f, 1.35f, 1.15f, 0.97f, 0.0f }, { ASTCENC_PRE_VERYTHOROUGH, - 4, 256, 128, 64, 98, 4, 6, 20, 14, 8, 200.0f, 200.0f, 10.0f, 1.6f, 1.4f, 0.98f + 4, 256, 128, 64, 98, 4, 6, 8, 6, 4, 200.0f, 200.0f, 10.0f, 1.6f, 1.4f, 0.98f, 0.0f }, { ASTCENC_PRE_EXHAUSTIVE, - 4, 512, 512, 512, 100, 4, 8, 32, 32, 32, 200.0f, 200.0f, 10.0f, 2.0f, 2.0f, 0.99f + 4, 512, 512, 512, 100, 4, 8, 8, 8, 8, 200.0f, 200.0f, 10.0f, 2.0f, 2.0f, 0.99f, 0.0f } }}; @@ -88,22 +89,22 @@ static const std::array<astcenc_preset_config, 6> preset_configs_high {{ static const std::array<astcenc_preset_config, 6> preset_configs_mid {{ { ASTCENC_PRE_FASTEST, - 2, 10, 6, 4, 43, 2, 2, 2, 2, 2, 85.2f, 63.2f, 3.5f, 1.0f, 1.0f, 0.80f + 2, 10, 6, 4, 43, 2, 2, 2, 2, 2, 85.2f, 63.2f, 3.5f, 1.0f, 1.0f, 0.80f, 1.0f }, { ASTCENC_PRE_FAST, - 3, 18, 12, 10, 55, 3, 3, 2, 2, 2, 85.2f, 63.2f, 3.5f, 1.0f, 1.0f, 0.85f + 3, 18, 12, 10, 55, 3, 3, 2, 2, 2, 85.2f, 63.2f, 3.5f, 1.0f, 1.0f, 0.85f, 1.0f }, { ASTCENC_PRE_MEDIUM, - 4, 34, 28, 16, 77, 3, 3, 2, 2, 2, 95.0f, 70.0f, 3.0f, 1.1f, 1.05f, 0.90f + 3, 34, 28, 16, 77, 3, 3, 2, 2, 2, 95.0f, 70.0f, 3.0f, 1.1f, 1.05f, 0.90f, 1.0f }, { ASTCENC_PRE_THOROUGH, - 4, 82, 60, 30, 94, 4, 4, 3, 2, 2, 105.0f, 77.0f, 10.0f, 1.4f, 1.2f, 0.95f + 4, 82, 60, 30, 94, 4, 4, 3, 2, 2, 105.0f, 77.0f, 10.0f, 1.4f, 1.2f, 0.95f, 0.0f }, { ASTCENC_PRE_VERYTHOROUGH, - 4, 256, 128, 64, 98, 4, 6, 12, 8, 3, 200.0f, 200.0f, 10.0f, 1.6f, 1.4f, 0.98f + 4, 256, 128, 64, 98, 4, 6, 8, 6, 3, 200.0f, 200.0f, 10.0f, 1.6f, 1.4f, 0.98f, 0.0f }, { ASTCENC_PRE_EXHAUSTIVE, - 4, 256, 256, 256, 100, 4, 8, 32, 32, 32, 200.0f, 200.0f, 10.0f, 2.0f, 2.0f, 0.99f + 4, 256, 256, 256, 100, 4, 8, 8, 8, 8, 200.0f, 200.0f, 10.0f, 2.0f, 2.0f, 0.99f, 0.0f } }}; @@ -113,22 +114,22 @@ static const std::array<astcenc_preset_config, 6> preset_configs_mid {{ static const std::array<astcenc_preset_config, 6> preset_configs_low {{ { ASTCENC_PRE_FASTEST, - 2, 10, 6, 4, 40, 2, 2, 2, 2, 2, 85.0f, 63.0f, 3.5f, 1.0f, 1.0f, 0.80f + 2, 10, 6, 4, 40, 2, 2, 2, 2, 2, 85.0f, 63.0f, 3.5f, 1.0f, 1.0f, 0.80f, 1.0f }, { ASTCENC_PRE_FAST, - 2, 18, 12, 10, 55, 3, 3, 2, 2, 2, 85.0f, 63.0f, 3.5f, 1.0f, 1.0f, 0.85f + 2, 18, 12, 10, 55, 3, 3, 2, 2, 2, 85.0f, 63.0f, 3.5f, 1.0f, 1.0f, 0.85f, 1.0f }, { ASTCENC_PRE_MEDIUM, - 3, 34, 28, 16, 77, 3, 3, 2, 2, 2, 95.0f, 70.0f, 3.5f, 1.1f, 1.05f, 0.90f + 3, 34, 28, 16, 77, 3, 3, 2, 2, 2, 95.0f, 70.0f, 3.5f, 1.1f, 1.05f, 0.90f, 1.0f }, { ASTCENC_PRE_THOROUGH, - 4, 82, 60, 30, 93, 4, 4, 3, 2, 2, 105.0f, 77.0f, 10.0f, 1.3f, 1.2f, 0.97f + 4, 82, 60, 30, 93, 4, 4, 3, 2, 2, 105.0f, 77.0f, 10.0f, 1.3f, 1.2f, 0.97f, 1.0f }, { ASTCENC_PRE_VERYTHOROUGH, - 4, 256, 128, 64, 98, 4, 6, 9, 5, 2, 200.0f, 200.0f, 10.0f, 1.6f, 1.4f, 0.98f + 4, 256, 128, 64, 98, 4, 6, 8, 5, 2, 200.0f, 200.0f, 10.0f, 1.6f, 1.4f, 0.98f, 1.0f }, { ASTCENC_PRE_EXHAUSTIVE, - 4, 256, 256, 256, 100, 4, 8, 32, 32, 32, 200.0f, 200.0f, 10.0f, 2.0f, 2.0f, 0.99f + 4, 256, 256, 256, 100, 4, 8, 8, 8, 8, 200.0f, 200.0f, 10.0f, 2.0f, 2.0f, 0.99f, 1.0f } }}; @@ -216,11 +217,13 @@ static astcenc_error validate_block_size( /** * @brief Validate flags. * - * @param flags The flags to check. + * @param profile The profile to check. + * @param flags The flags to check. * * @return Return @c ASTCENC_SUCCESS if validated, otherwise an error on failure. */ static astcenc_error validate_flags( + astcenc_profile profile, unsigned int flags ) { // Flags field must not contain any unknown flag bits @@ -238,6 +241,14 @@ static astcenc_error validate_flags( return ASTCENC_ERR_BAD_FLAGS; } + // Decode_unorm8 must only be used with an LDR profile + bool is_unorm8 = flags & ASTCENC_FLG_USE_DECODE_UNORM8; + bool is_hdr = (profile == ASTCENC_PRF_HDR) || (profile == ASTCENC_PRF_HDR_RGB_LDR_A); + if (is_unorm8 && is_hdr) + { + return ASTCENC_ERR_BAD_DECODE_MODE; + } + return ASTCENC_SUCCESS; } @@ -363,7 +374,7 @@ static astcenc_error validate_config( return status; } - status = validate_flags(config.flags); + status = validate_flags(config.profile, config.flags); if (status != ASTCENC_SUCCESS) { return status; @@ -504,10 +515,10 @@ astcenc_error astcenc_config_init( config.tune_4partition_index_limit = (*preset_configs)[start].tune_4partition_index_limit; config.tune_block_mode_limit = (*preset_configs)[start].tune_block_mode_limit; config.tune_refinement_limit = (*preset_configs)[start].tune_refinement_limit; - config.tune_candidate_limit = astc::min((*preset_configs)[start].tune_candidate_limit, TUNE_MAX_TRIAL_CANDIDATES); - config.tune_2partitioning_candidate_limit = astc::min((*preset_configs)[start].tune_2partitioning_candidate_limit, TUNE_MAX_PARTITIONING_CANDIDATES); - config.tune_3partitioning_candidate_limit = astc::min((*preset_configs)[start].tune_3partitioning_candidate_limit, TUNE_MAX_PARTITIONING_CANDIDATES); - config.tune_4partitioning_candidate_limit = astc::min((*preset_configs)[start].tune_4partitioning_candidate_limit, TUNE_MAX_PARTITIONING_CANDIDATES); + config.tune_candidate_limit = (*preset_configs)[start].tune_candidate_limit; + config.tune_2partitioning_candidate_limit = (*preset_configs)[start].tune_2partitioning_candidate_limit; + config.tune_3partitioning_candidate_limit = (*preset_configs)[start].tune_3partitioning_candidate_limit; + config.tune_4partitioning_candidate_limit = (*preset_configs)[start].tune_4partitioning_candidate_limit; config.tune_db_limit = astc::max((*preset_configs)[start].tune_db_limit_a_base - 35 * ltexels, (*preset_configs)[start].tune_db_limit_b_base - 19 * ltexels); @@ -516,6 +527,7 @@ astcenc_error astcenc_config_init( config.tune_2partition_early_out_limit_factor = (*preset_configs)[start].tune_2partition_early_out_limit_factor; config.tune_3partition_early_out_limit_factor = (*preset_configs)[start].tune_3partition_early_out_limit_factor; config.tune_2plane_early_out_limit_correlation = (*preset_configs)[start].tune_2plane_early_out_limit_correlation; + config.tune_search_mode0_enable = (*preset_configs)[start].tune_search_mode0_enable; } // Start and end node are not the same - so interpolate between them else @@ -542,14 +554,10 @@ astcenc_error astcenc_config_init( config.tune_4partition_index_limit = LERPI(tune_4partition_index_limit); config.tune_block_mode_limit = LERPI(tune_block_mode_limit); config.tune_refinement_limit = LERPI(tune_refinement_limit); - config.tune_candidate_limit = astc::min(LERPUI(tune_candidate_limit), - TUNE_MAX_TRIAL_CANDIDATES); - config.tune_2partitioning_candidate_limit = astc::min(LERPUI(tune_2partitioning_candidate_limit), - BLOCK_MAX_PARTITIONINGS); - config.tune_3partitioning_candidate_limit = astc::min(LERPUI(tune_3partitioning_candidate_limit), - BLOCK_MAX_PARTITIONINGS); - config.tune_4partitioning_candidate_limit = astc::min(LERPUI(tune_4partitioning_candidate_limit), - BLOCK_MAX_PARTITIONINGS); + config.tune_candidate_limit = LERPUI(tune_candidate_limit); + config.tune_2partitioning_candidate_limit = LERPUI(tune_2partitioning_candidate_limit); + config.tune_3partitioning_candidate_limit = LERPUI(tune_3partitioning_candidate_limit); + config.tune_4partitioning_candidate_limit = LERPUI(tune_4partitioning_candidate_limit); config.tune_db_limit = astc::max(LERP(tune_db_limit_a_base) - 35 * ltexels, LERP(tune_db_limit_b_base) - 19 * ltexels); @@ -558,6 +566,7 @@ astcenc_error astcenc_config_init( config.tune_2partition_early_out_limit_factor = LERP(tune_2partition_early_out_limit_factor); config.tune_3partition_early_out_limit_factor = LERP(tune_3partition_early_out_limit_factor); config.tune_2plane_early_out_limit_correlation = LERP(tune_2plane_early_out_limit_correlation); + config.tune_search_mode0_enable = LERP(tune_search_mode0_enable); #undef LERP #undef LERPI #undef LERPUI @@ -585,13 +594,14 @@ astcenc_error astcenc_config_init( case ASTCENC_PRF_HDR_RGB_LDR_A: case ASTCENC_PRF_HDR: config.tune_db_limit = 999.0f; + config.tune_search_mode0_enable = 0.0f; break; default: return ASTCENC_ERR_BAD_PROFILE; } // Flags field must not contain any unknown flag bits - status = validate_flags(flags); + status = validate_flags(profile, flags); if (status != ASTCENC_SUCCESS) { return status; @@ -689,6 +699,12 @@ astcenc_error astcenc_context_alloc( } ctx->bsd = aligned_malloc<block_size_descriptor>(sizeof(block_size_descriptor), ASTCENC_VECALIGN); + if (!ctx->bsd) + { + delete ctxo; + return ASTCENC_ERR_OUT_OF_MEM; + } + bool can_omit_modes = static_cast<bool>(config.flags & ASTCENC_FLG_SELF_DECOMPRESS_ONLY); init_block_size_descriptor(config.block_x, config.block_y, config.block_z, can_omit_modes, @@ -698,7 +714,7 @@ astcenc_error astcenc_context_alloc( #if !defined(ASTCENC_DECOMPRESS_ONLY) // Do setup only needed by compression - if (!(status & ASTCENC_FLG_DECOMPRESS_ONLY)) + if (!(ctx->config.flags & ASTCENC_FLG_DECOMPRESS_ONLY)) { // Turn a dB limit into a per-texel error for faster use later if ((ctx->config.profile == ASTCENC_PRF_LDR) || (ctx->config.profile == ASTCENC_PRF_LDR_SRGB)) @@ -712,7 +728,7 @@ astcenc_error astcenc_context_alloc( size_t worksize = sizeof(compression_working_buffers) * thread_count; ctx->working_buffers = aligned_malloc<compression_working_buffers>(worksize, ASTCENC_VECALIGN); - static_assert((sizeof(compression_working_buffers) % ASTCENC_VECALIGN) == 0, + static_assert((ASTCENC_VECALIGN == 0) || ((sizeof(compression_working_buffers) % ASTCENC_VECALIGN) == 0), "compression_working_buffers size must be multiple of vector alignment"); if (!ctx->working_buffers) { @@ -802,6 +818,8 @@ static void compress_image( int row_blocks = xblocks; int plane_blocks = xblocks * yblocks; + blk.decode_unorm8 = ctxo.context.config.flags & ASTCENC_FLG_USE_DECODE_UNORM8; + // Populate the block channel weights blk.channel_weight = vfloat4(ctx.config.cw_r_weight, ctx.config.cw_g_weight, @@ -812,7 +830,7 @@ static void compress_image( auto& temp_buffers = ctx.working_buffers[thread_index]; // Only the first thread actually runs the initializer - ctxo.manage_compress.init(block_count); + ctxo.manage_compress.init(block_count, ctx.config.progress_callback); // Determine if we can use an optimized load function bool needs_swz = (swizzle.r != ASTCENC_SWZ_R) || (swizzle.g != ASTCENC_SWZ_G) || @@ -914,8 +932,7 @@ static void compress_image( int offset = ((z * yblocks + y) * xblocks + x) * 16; uint8_t *bp = buffer + offset; - physical_compressed_block* pcb = reinterpret_cast<physical_compressed_block*>(bp); - compress_block(ctx, blk, *pcb, temp_buffers); + compress_block(ctx, blk, bp, temp_buffers); } ctxo.manage_compress.complete_task_assignment(count); @@ -1138,6 +1155,7 @@ astcenc_error astcenc_decompress_image( unsigned int xblocks = (image_out.dim_x + block_x - 1) / block_x; unsigned int yblocks = (image_out.dim_y + block_y - 1) / block_y; unsigned int zblocks = (image_out.dim_z + block_z - 1) / block_z; + unsigned int block_count = zblocks * yblocks * xblocks; int row_blocks = xblocks; int plane_blocks = xblocks * yblocks; @@ -1152,6 +1170,9 @@ astcenc_error astcenc_decompress_image( image_block blk; blk.texel_count = static_cast<uint8_t>(block_x * block_y * block_z); + // Decode mode inferred from the output data type + blk.decode_unorm8 = image_out.data_type == ASTCENC_TYPE_U8; + // If context thread count is one then implicitly reset if (ctx->thread_count == 1) { @@ -1159,7 +1180,7 @@ astcenc_error astcenc_decompress_image( } // Only the first thread actually runs the initializer - ctxo->manage_decompress.init(zblocks * yblocks * xblocks); + ctxo->manage_decompress.init(block_count, nullptr); // All threads run this processing loop until there is no work remaining while (true) @@ -1182,10 +1203,9 @@ astcenc_error astcenc_decompress_image( unsigned int offset = (((z * yblocks + y) * xblocks) + x) * 16; const uint8_t* bp = data + offset; - const physical_compressed_block& pcb = *reinterpret_cast<const physical_compressed_block*>(bp); symbolic_compressed_block scb; - physical_to_symbolic(*ctx->bsd, pcb, scb); + physical_to_symbolic(*ctx->bsd, bp, scb); decompress_symbolic_block(ctx->config.profile, *ctx->bsd, x * block_x, y * block_y, z * block_z, @@ -1224,9 +1244,8 @@ astcenc_error astcenc_get_block_info( astcenc_contexti* ctx = &ctxo->context; // Decode the compressed data into a symbolic form - const physical_compressed_block&pcb = *reinterpret_cast<const physical_compressed_block*>(data); symbolic_compressed_block scb; - physical_to_symbolic(*ctx->bsd, pcb, scb); + physical_to_symbolic(*ctx->bsd, data, scb); // Fetch the appropriate partition and decimation tables block_size_descriptor& bsd = *ctx->bsd; @@ -1359,6 +1378,8 @@ const char* astcenc_get_error_string( return "ASTCENC_ERR_BAD_CONTEXT"; case ASTCENC_ERR_NOT_IMPLEMENTED: return "ASTCENC_ERR_NOT_IMPLEMENTED"; + case ASTCENC_ERR_BAD_DECODE_MODE: + return "ASTCENC_ERR_BAD_DECODE_MODE"; #if defined(ASTCENC_DIAGNOSTICS) case ASTCENC_ERR_DTRACE_FAILURE: return "ASTCENC_ERR_DTRACE_FAILURE"; |