lookback param in blocks

Author: YunHsiao

Source/astcenc.h

@@ -587,9 +587,9 @@ struct astcenc_config
 	float rdo_quality;
 
 	/**
-	 * @brief RDO dictionary size in bytes.
+	 * @brief RDO lookback size in blocks.
 	 */
-	unsigned int rdo_dict_size;
+	unsigned int rdo_lookback;
 
 	/**
 	 * @brief RDO task partitions.

Source/astcenc_entry.cpp

@@ -56,8 +56,7 @@ struct astcenc_preset_config
 	float tune_3partition_early_out_limit_factor;
 	float tune_2plane_early_out_limit_correlation;
 	float tune_search_mode0_enable;
-	float rdo_quality;
-	unsigned int rdo_dict_size;
+	unsigned int rdo_lookback;
 };
 
 /**
@@ -66,22 +65,22 @@ 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, 0.0f, 4.0f, 256
+		2, 10, 6, 4, 43, 2, 2, 2, 2, 2, 85.2f, 63.2f, 3.5f, 1.0f, 1.0f, 0.85f, 0.0f, 64
 	}, {
 		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, 0.0f, 2.0f, 1024
+		3, 18, 10, 8, 55, 3, 3, 2, 2, 2, 85.2f, 63.2f, 3.5f, 1.0f, 1.0f, 0.90f, 0.0f, 128
 	}, {
 		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, 0.0f, 1.0f, 4096
+		4, 34, 28, 16, 77, 3, 3, 2, 2, 2, 95.0f, 70.0f, 2.5f, 1.1f, 1.05f, 0.95f, 0.0f, 256
 	}, {
 		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, 0.0f, 0.5f, 4096
+		4, 82, 60, 30, 94, 4, 4, 3, 2, 2, 105.0f, 77.0f, 10.0f, 1.35f, 1.15f, 0.97f, 0.0f, 256
 	}, {
 		ASTCENC_PRE_VERYTHOROUGH,
-		4, 256, 128, 64, 98, 4, 6, 8, 6, 4, 200.0f, 200.0f, 10.0f, 1.6f, 1.4f, 0.98f, 0.0f, 0.4f, 4096
+		4, 256, 128, 64, 98, 4, 6, 8, 6, 4, 200.0f, 200.0f, 10.0f, 1.6f, 1.4f, 0.98f, 0.0f, 256
 	}, {
 		ASTCENC_PRE_EXHAUSTIVE,
-		4, 512, 512, 512, 100, 4, 8, 8, 8, 8, 200.0f, 200.0f, 10.0f, 2.0f, 2.0f, 0.99f, 0.0f, 0.2f, 4096
+		4, 512, 512, 512, 100, 4, 8, 8, 8, 8, 200.0f, 200.0f, 10.0f, 2.0f, 2.0f, 0.99f, 0.0f, 256
 	}
 }};
 
@@ -91,22 +90,22 @@ 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, 1.0f, 4.0f, 256
+		2, 10, 6, 4, 43, 2, 2, 2, 2, 2, 85.2f, 63.2f, 3.5f, 1.0f, 1.0f, 0.80f, 1.0f, 64
 	}, {
 		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, 1.0f, 2.0f, 1024
+		3, 18, 12, 10, 55, 3, 3, 2, 2, 2, 85.2f, 63.2f, 3.5f, 1.0f, 1.0f, 0.85f, 1.0f, 128
 	}, {
 		ASTCENC_PRE_MEDIUM,
-		3, 34, 28, 16, 77, 3, 3, 2, 2, 2, 95.0f, 70.0f, 3.0f, 1.1f, 1.05f, 0.90f, 1.0f, 1.0f, 4096
+		3, 34, 28, 16, 77, 3, 3, 2, 2, 2, 95.0f, 70.0f, 3.0f, 1.1f, 1.05f, 0.90f, 1.0f, 256
 	}, {
 		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, 0.0f, 0.5f, 4096
+		4, 82, 60, 30, 94, 4, 4, 3, 2, 2, 105.0f, 77.0f, 10.0f, 1.4f, 1.2f, 0.95f, 0.0f, 256
 	}, {
 		ASTCENC_PRE_VERYTHOROUGH,
-		4, 256, 128, 64, 98, 4, 6, 8, 6, 3, 200.0f, 200.0f, 10.0f, 1.6f, 1.4f, 0.98f, 0.0f, 0.4f, 4096
+		4, 256, 128, 64, 98, 4, 6, 8, 6, 3, 200.0f, 200.0f, 10.0f, 1.6f, 1.4f, 0.98f, 0.0f, 256
 	}, {
 		ASTCENC_PRE_EXHAUSTIVE,
-		4, 256, 256, 256, 100, 4, 8, 8, 8, 8, 200.0f, 200.0f, 10.0f, 2.0f, 2.0f, 0.99f, 0.0f, 0.2f, 4096
+		4, 256, 256, 256, 100, 4, 8, 8, 8, 8, 200.0f, 200.0f, 10.0f, 2.0f, 2.0f, 0.99f, 0.0f, 256
 	}
 }};
 
@@ -116,22 +115,22 @@ 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, 1.0f, 4.0f, 256
+		2, 10, 6, 4, 40, 2, 2, 2, 2, 2, 85.0f, 63.0f, 3.5f, 1.0f, 1.0f, 0.80f, 1.0f, 64
 	}, {
 		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, 1.0f, 2.0f, 1024
+		2, 18, 12, 10, 55, 3, 3, 2, 2, 2, 85.0f, 63.0f, 3.5f, 1.0f, 1.0f, 0.85f, 1.0f, 128
 	}, {
 		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, 1.0f, 1.0f, 4096
+		3, 34, 28, 16, 77, 3, 3, 2, 2, 2, 95.0f, 70.0f, 3.5f, 1.1f, 1.05f, 0.90f, 1.0f, 256
 	}, {
 		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, 1.0f, 0.5f, 4096
+		4, 82, 60, 30, 93, 4, 4, 3, 2, 2, 105.0f, 77.0f, 10.0f, 1.3f, 1.2f, 0.97f, 1.0f, 256
 	}, {
 		ASTCENC_PRE_VERYTHOROUGH,
-		4, 256, 128, 64, 98, 4, 6, 8, 5, 2, 200.0f, 200.0f, 10.0f, 1.6f, 1.4f, 0.98f, 1.0f, 0.4f, 4096
+		4, 256, 128, 64, 98, 4, 6, 8, 5, 2, 200.0f, 200.0f, 10.0f, 1.6f, 1.4f, 0.98f, 1.0f, 256
 	}, {
 		ASTCENC_PRE_EXHAUSTIVE,
-		4, 256, 256, 256, 100, 4, 8, 8, 8, 8, 200.0f, 200.0f, 10.0f, 2.0f, 2.0f, 0.99f, 1.0f, 0.2f, 4096
+		4, 256, 256, 256, 100, 4, 8, 8, 8, 8, 200.0f, 200.0f, 10.0f, 2.0f, 2.0f, 0.99f, 1.0f, 256
 	}
 }};
 
@@ -415,7 +414,7 @@ static astcenc_error validate_config(
 	config.tune_2plane_early_out_limit_correlation = astc::max(config.tune_2plane_early_out_limit_correlation, 0.0f);
 
 	config.rdo_quality = astc::clamp(config.rdo_quality, 0.001f, 50.0f);
-	config.rdo_dict_size = astc::clamp(config.rdo_dict_size, 64u, 65536u);
+	config.rdo_lookback = astc::clamp(config.rdo_lookback, 4u, 4096u);
 	config.rdo_max_smooth_block_error_scale = astc::clamp(config.rdo_max_smooth_block_error_scale, 1.0f, 300.0f);
 	config.rdo_max_smooth_block_std_dev = astc::clamp(config.rdo_max_smooth_block_std_dev, 0.01f, 65536.0f);
 
@@ -535,8 +534,7 @@ astcenc_error astcenc_config_init(
 		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;
-		config.rdo_quality = (*preset_configs)[start].rdo_quality;
-		config.rdo_dict_size = (*preset_configs)[start].rdo_dict_size;
+		config.rdo_lookback = (*preset_configs)[start].rdo_lookback;
 	}
 	// Start and end node are not the same - so interpolate between them
 	else
@@ -576,13 +574,13 @@ astcenc_error astcenc_config_init(
 		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);
-		config.rdo_quality = LERP(rdo_quality);
-		config.rdo_dict_size = LERPUI(rdo_dict_size);
+		config.rdo_lookback = LERPUI(rdo_lookback);
 		#undef LERP
 		#undef LERPI
 		#undef LERPUI
 	}
 
+	config.rdo_quality = 0.5f;
 	config.rdo_max_smooth_block_error_scale = 10.0f;
 	config.rdo_max_smooth_block_std_dev = 18.0f;
 

Source/astcenc_rate_distortion.cpp

@@ -92,7 +92,7 @@ static uint32_t init_rdo_context(
 	// Generate quality parameters
 	auto& ert_params = rdo_ctx.m_ert_params;
 	ert_params.m_lambda = ctx.config.rdo_quality;
-	ert_params.m_lookback_window_size = ctx.config.rdo_dict_size;
+	ert_params.m_lookback_window_size = ctx.config.rdo_lookback * ASTCENC_BYTES_PER_BLOCK;
 	ert_params.m_smooth_block_max_mse_scale = ctx.config.rdo_max_smooth_block_error_scale;
 	ert_params.m_max_smooth_block_std_dev = ctx.config.rdo_max_smooth_block_std_dev;
 
@@ -422,7 +422,7 @@ void rate_distortion_optimize(
 	uint32_t blocks_per_task = total_blocks;
 	if (!ctx.config.rdo_no_multithreading)
 	{
-		blocks_per_task = astc::min(ctx.config.rdo_dict_size / ASTCENC_BYTES_PER_BLOCK, total_blocks);
+		blocks_per_task = astc::min(ctx.config.rdo_lookback, total_blocks);
 		// There is no way to losslessly partition the job (sequentially dependent on previous output)
 		// So we reserve only one task for each thread to minimize the quality impact.
 		uint32_t partitions = ctx.config.rdo_partitions ? ctx.config.rdo_partitions : ctx.thread_count;

Source/astcenccli_toplevel.cpp

@@ -1179,12 +1179,6 @@ static int edit_astcenc_config(
 		{
 			argidx += 1;
 			config.rdo_enabled = true;
-
-			// Unpacking RDO trials blocks requires full initialization
-			if (static_cast<bool>(config.flags & ASTCENC_FLG_SELF_DECOMPRESS_ONLY))
-			{
-				config.flags &= ~ASTCENC_FLG_SELF_DECOMPRESS_ONLY;
-			}
 		}
 		else if (!strcmp(argv[argidx], "-rdo-no-multithreading"))
 		{
@@ -1200,6 +1194,7 @@ static int edit_astcenc_config(
 				return 1;
 			}
 
+			config.rdo_enabled = true;
 			config.rdo_quality = static_cast<float>(atof(argv[argidx - 1]));
 		}
 		else if (!strcmp(argv[argidx], "-rdo-dict-size"))
@@ -1211,7 +1206,20 @@ static int edit_astcenc_config(
 				return 1;
 			}
 
-			config.rdo_dict_size = atoi(argv[argidx - 1]);
+			config.rdo_enabled = true;
+			config.rdo_lookback = atoi(argv[argidx - 1]) / 16;
+		}
+		else if (!strcmp(argv[argidx], "-rdo-lookback"))
+		{
+			argidx += 2;
+			if (argidx > argc)
+			{
+				print_error("ERROR: -rdo-dict-size switch with no argument\n");
+				return 1;
+			}
+
+			config.rdo_enabled = true;
+			config.rdo_lookback = atoi(argv[argidx - 1]);
 		}
 		else if (!strcmp(argv[argidx], "-rdo-partitions"))
 		{
@@ -1222,6 +1230,7 @@ static int edit_astcenc_config(
 				return 1;
 			}
 
+			config.rdo_enabled = true;
 			config.rdo_partitions = atoi(argv[argidx - 1]);
 		}
 		else if (!strcmp(argv[argidx], "-rdo-max-smooth-block-error-scale"))
@@ -1233,6 +1242,7 @@ static int edit_astcenc_config(
 				return 1;
 			}
 
+			config.rdo_enabled = true;
 			config.rdo_max_smooth_block_error_scale = static_cast<float>(atof(argv[argidx - 1]));
 		}
 		else if (!strcmp(argv[argidx], "-rdo-max-smooth-block-std-dev"))
@@ -1244,6 +1254,7 @@ static int edit_astcenc_config(
 				return 1;
 			}
 
+			config.rdo_enabled = true;
 			config.rdo_max_smooth_block_std_dev = static_cast<float>(atof(argv[argidx - 1]));
 		}
 		else // check others as well
@@ -1341,10 +1352,10 @@ static void print_astcenc_config(
 		{
 			printf("    RDO multithreading:         %s\n", config.rdo_no_multithreading ? "Disabled" : "Enabled");
 			printf("    RDO quality:                %g\n", static_cast<double>(config.rdo_quality));
-			printf("    RDO dictionary size:        %u bytes\n", config.rdo_dict_size);
+			printf("    RDO lookback:               %u blocks\n", config.rdo_lookback);
 			printf("    RDO max error scale:        %g\n", static_cast<double>(config.rdo_max_smooth_block_error_scale));
 			printf("    RDO max standard deviation: %g\n", static_cast<double>(config.rdo_max_smooth_block_std_dev));
-			if (config.rdo_partitions) printf("    RDO partitions:             %u\n", cli_config.thread_count);
+			if (config.rdo_partitions) printf("    RDO partitions:             %u\n", config.rdo_partitions);
 		}
 		printf("\n");
 	}
@@ -2070,6 +2081,12 @@ int astcenc_main(
 		return 1;
 	}
 
+	// Unpacking RDO trials blocks requires full initialization
+	if (config.rdo_enabled && static_cast<bool>(config.flags & ASTCENC_FLG_SELF_DECOMPRESS_ONLY))
+	{
+		config.flags &= ~ASTCENC_FLG_SELF_DECOMPRESS_ONLY;
+	}
+
 	// Enable progress callback if not in silent mode and using a terminal
 	#if defined(_WIN32)
 		int stdoutfno = _fileno(stdout);

Source/astcenccli_toplevel_help.cpp

@@ -396,25 +396,21 @@ ADVANCED COMPRESSION
            RDO quality scalar (lambda). Lower values yield higher
            quality/larger LZ compressed files, higher values yield lower
            quality/smaller LZ compressed files. A good range to try is [.2,4].
-           Full range is [.001,50.0]. Preset defaults are:
+           Full range is [.001,50.0]. Default to 0.5.
 
-               -fastest      : 4.0
-               -fast         : 2.0
-               -medium       : 1.0
-               -thorough     : 0.5
-               -verythorough : 0.4
-               -exhaustive   : 0.2
+       -rdo-lookback <number>
+           RDO look back size in blocks. Lower values=faster,
+           but give less compression. Range is [4,4096]. Preset defaults are:
+
+               -fastest      : 64
+               -fast         : 128
+               -medium       : 256
+               -thorough     : 256
+               -verythorough : 256
+               -exhaustive   : 256
 
        -rdo-dict-size <number>
-           RDO dictionary size in bytes. Lower values=faster,
-           but give less compression. Range is [64,65536]. Preset defaults are:
-
-               -fastest      : 256
-               -fast         : 1024
-               -medium       : 4096
-               -thorough     : 4096
-               -verythorough : 4096
-               -exhaustive   : 4096
+           Same as rdo-lookback, but in bytes.
 
        -rdo-partitions <number>
            RDO task partitions. Default to current number of threads.