提交: 9259c64445832fc2f3412b0a37b5be396376d1e6 - color_quantizer (git)

減色プログラム

Commit MetaInfo

• 修订版: 9259c64445832fc2f3412b0a37b5be396376d1e6 (tree)
• 时间: 2011-05-22 02:25:58
• 作者: berupon <berupon@gmai...>
• Commiter: berupon

Log Message

optimized
added AMD64 platform

更改概述

• modified: Color4d.h
• modified: color_quantizer.sln
• modified: color_quantizer.vcproj
• modified: main.cpp
• modified: quantize.cpp
• modified: stdafx.h

差异

--- a/Color4d.h
+++ b/Color4d.h
@@ -101,22 +101,12 @@ struct Color4d
 	}
 
 	double& operator[] (int idx) {
-		switch (idx) {
-		case 0: return v[0].m128d_f64[0];
-		case 1: return v[0].m128d_f64[1];
-		case 2: return v[1].m128d_f64[0];
-		case 3: return v[1].m128d_f64[1];
-		}
+		return ((double*)&v)[idx];
 	}
 	const double& operator[] (int idx) const {
-		switch (idx) {
-		case 0: return v[0].m128d_f64[0];
-		case 1: return v[0].m128d_f64[1];
-		case 2: return v[1].m128d_f64[0];
-		case 3: return v[1].m128d_f64[1];
-		}
+		return ((double*)&v)[idx];
 	}
-
+	
 	double norm_squared() {
 		double result = 0;
 		for (int i=0; i<3; i++) {

--- a/color_quantizer.sln
+++ b/color_quantizer.sln
@@ -6,20 +6,26 @@ EndProject
 Global
 	GlobalSection(SolutionConfigurationPlatforms) = preSolution
 		Debug|Win32 = Debug|Win32
+		Debug|x64 = Debug|x64
 		Release|Win32 = Release|Win32
+		Release|x64 = Release|x64
 	EndGlobalSection
 	GlobalSection(ProjectConfigurationPlatforms) = postSolution
 		{21F9C24D-7A00-425D-B59C-E1CD45C445A6}.Debug|Win32.ActiveCfg = Debug|Win32
 		{21F9C24D-7A00-425D-B59C-E1CD45C445A6}.Debug|Win32.Build.0 = Debug|Win32
+		{21F9C24D-7A00-425D-B59C-E1CD45C445A6}.Debug|x64.ActiveCfg = Debug|x64
+		{21F9C24D-7A00-425D-B59C-E1CD45C445A6}.Debug|x64.Build.0 = Debug|x64
 		{21F9C24D-7A00-425D-B59C-E1CD45C445A6}.Release|Win32.ActiveCfg = Release|Win32
 		{21F9C24D-7A00-425D-B59C-E1CD45C445A6}.Release|Win32.Build.0 = Release|Win32
+		{21F9C24D-7A00-425D-B59C-E1CD45C445A6}.Release|x64.ActiveCfg = Release|x64
+		{21F9C24D-7A00-425D-B59C-E1CD45C445A6}.Release|x64.Build.0 = Release|x64
 	EndGlobalSection
 	GlobalSection(SolutionProperties) = preSolution
 		HideSolutionNode = FALSE
 	EndGlobalSection
 	GlobalSection(ExtensibilityGlobals) = postSolution
-		AMDCaPersistentConfig = Release|Win32
-		AMDCaPersistentStartup = color_quantizer
 		AMDCaProjectFile = C:\projects\color_quantizer\CodeAnalyst\color_quantizer.caw
+		AMDCaPersistentStartup = color_quantizer
+		AMDCaPersistentConfig = Release|Win32
 	EndGlobalSection
 EndGlobal

--- a/color_quantizer.vcproj
+++ b/color_quantizer.vcproj
@@ -12,6 +12,9 @@
 		<Platform
 			Name="Win32"
 		/>
+		<Platform
+			Name="x64"
+		/>
 	</Platforms>
 	<ToolFiles>
 	</ToolFiles>
@@ -170,6 +173,162 @@
 				Name="VCPostBuildEventTool"
 			/>
 		</Configuration>
+		<Configuration
+			Name="Debug|x64"
+			OutputDirectory="$(SolutionDir)$(PlatformName)\$(ConfigurationName)"
+			IntermediateDirectory="$(PlatformName)\$(ConfigurationName)"
+			ConfigurationType="1"
+			CharacterSet="1"
+			>
+			<Tool
+				Name="VCPreBuildEventTool"
+			/>
+			<Tool
+				Name="VCCustomBuildTool"
+			/>
+			<Tool
+				Name="VCXMLDataGeneratorTool"
+			/>
+			<Tool
+				Name="VCWebServiceProxyGeneratorTool"
+			/>
+			<Tool
+				Name="VCMIDLTool"
+				TargetEnvironment="3"
+			/>
+			<Tool
+				Name="VCCLCompilerTool"
+				Optimization="0"
+				AdditionalIncludeDirectories="./"
+				PreprocessorDefinitions="WIN32;_DEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS"
+				MinimalRebuild="true"
+				BasicRuntimeChecks="3"
+				RuntimeLibrary="3"
+				UsePrecompiledHeader="2"
+				WarningLevel="3"
+				DebugInformationFormat="3"
+				ForcedIncludeFiles="common.h"
+			/>
+			<Tool
+				Name="VCManagedResourceCompilerTool"
+			/>
+			<Tool
+				Name="VCResourceCompilerTool"
+			/>
+			<Tool
+				Name="VCPreLinkEventTool"
+			/>
+			<Tool
+				Name="VCLinkerTool"
+				LinkIncremental="2"
+				GenerateDebugInformation="true"
+				SubSystem="1"
+				TargetMachine="17"
+			/>
+			<Tool
+				Name="VCALinkTool"
+			/>
+			<Tool
+				Name="VCManifestTool"
+			/>
+			<Tool
+				Name="VCXDCMakeTool"
+			/>
+			<Tool
+				Name="VCBscMakeTool"
+			/>
+			<Tool
+				Name="VCFxCopTool"
+			/>
+			<Tool
+				Name="VCAppVerifierTool"
+			/>
+			<Tool
+				Name="VCPostBuildEventTool"
+			/>
+		</Configuration>
+		<Configuration
+			Name="Release|x64"
+			OutputDirectory="$(SolutionDir)$(PlatformName)\$(ConfigurationName)"
+			IntermediateDirectory="$(PlatformName)\$(ConfigurationName)"
+			ConfigurationType="1"
+			CharacterSet="1"
+			WholeProgramOptimization="1"
+			>
+			<Tool
+				Name="VCPreBuildEventTool"
+			/>
+			<Tool
+				Name="VCCustomBuildTool"
+			/>
+			<Tool
+				Name="VCXMLDataGeneratorTool"
+			/>
+			<Tool
+				Name="VCWebServiceProxyGeneratorTool"
+			/>
+			<Tool
+				Name="VCMIDLTool"
+				TargetEnvironment="3"
+			/>
+			<Tool
+				Name="VCCLCompilerTool"
+				Optimization="2"
+				EnableIntrinsicFunctions="true"
+				FavorSizeOrSpeed="1"
+				AdditionalIncludeDirectories="./"
+				PreprocessorDefinitions="WIN32;NDEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS"
+				RuntimeLibrary="0"
+				BufferSecurityCheck="false"
+				EnableFunctionLevelLinking="true"
+				FloatingPointModel="2"
+				UsePrecompiledHeader="2"
+				BrowseInformation="1"
+				WarningLevel="3"
+				DebugInformationFormat="3"
+				ForcedIncludeFiles="common.h"
+			/>
+			<Tool
+				Name="VCManagedResourceCompilerTool"
+			/>
+			<Tool
+				Name="VCResourceCompilerTool"
+			/>
+			<Tool
+				Name="VCPreLinkEventTool"
+			/>
+			<Tool
+				Name="VCLinkerTool"
+				LinkIncremental="1"
+				GenerateDebugInformation="true"
+				SubSystem="1"
+				OptimizeReferences="2"
+				EnableCOMDATFolding="2"
+				RandomizedBaseAddress="1"
+				TargetMachine="17"
+			/>
+			<Tool
+				Name="VCALinkTool"
+			/>
+			<Tool
+				Name="VCManifestTool"
+			/>
+			<Tool
+				Name="VCXDCMakeTool"
+			/>
+			<Tool
+				Name="VCBscMakeTool"
+			/>
+			<Tool
+				Name="VCFxCopTool"
+			/>
+			<Tool
+				Name="VCAppVerifierTool"
+			/>
+			<Tool
+				Name="VCPostBuildEventTool"
+			/>
+		</Configuration>
 	</Configurations>
 	<References>
 	</References>
@@ -214,6 +373,22 @@
 						UsePrecompiledHeader="1"
 					/>
 				</FileConfiguration>
+				<FileConfiguration
+					Name="Debug|x64"
+					>
+					<Tool
+						Name="VCCLCompilerTool"
+						UsePrecompiledHeader="1"
+					/>
+				</FileConfiguration>
+				<FileConfiguration
+					Name="Release|x64"
+					>
+					<Tool
+						Name="VCCLCompilerTool"
+						UsePrecompiledHeader="1"
+					/>
+				</FileConfiguration>
 			</File>
 		</Filter>
 		<Filter
@@ -280,6 +455,22 @@
 						UsePrecompiledHeader="0"
 					/>
 				</FileConfiguration>
+				<FileConfiguration
+					Name="Debug|x64"
+					>
+					<Tool
+						Name="VCCLCompilerTool"
+						UsePrecompiledHeader="0"
+					/>
+				</FileConfiguration>
+				<FileConfiguration
+					Name="Release|x64"
+					>
+					<Tool
+						Name="VCCLCompilerTool"
+						UsePrecompiledHeader="0"
+					/>
+				</FileConfiguration>
 			</File>
 			<File
 				RelativePath=".\ReadImage\File.h"
@@ -308,6 +499,22 @@
 						UsePrecompiledHeader="0"
 					/>
 				</FileConfiguration>
+				<FileConfiguration
+					Name="Debug|x64"
+					>
+					<Tool
+						Name="VCCLCompilerTool"
+						UsePrecompiledHeader="0"
+					/>
+				</FileConfiguration>
+				<FileConfiguration
+					Name="Release|x64"
+					>
+					<Tool
+						Name="VCCLCompilerTool"
+						UsePrecompiledHeader="0"
+					/>
+				</FileConfiguration>
 			</File>
 			<File
 				RelativePath=".\ReadImage\ReadImage.h"

--- a/main.cpp
+++ b/main.cpp
@@ -144,15 +144,13 @@ int _tmain(int argc, _TCHAR* argv[])
 		for (int j=0; j<3; j++) {
 			double w = exp(-sqrt((double)((i-1)*(i-1) + (j-1)*(j-1)))/(stddev*stddev));
 			filter3_weights[i][j] = Color(w,w,w,0);
-			sum += 3 * w;
+			sum += w;
 		}
 	}
-	sum /= 3;
+	double invSum = 1.0 / sum;
 	for (int i=0; i<3; i++) {
 		for (int j=0; j<3; j++) {
-			for (int k=0; k<3; k++) {
-				filter3_weights[i][j][k] /= sum;
-			}
+			filter3_weights[i][j] *= invSum;
 		}
 	}
 	sum = 0.0;
@@ -160,15 +158,13 @@ int _tmain(int argc, _TCHAR* argv[])
 		for (int j=0; j<5; j++) {
 			double w = exp(-sqrt((double)((i-2)*(i-2) + (j-2)*(j-2)))/(stddev*stddev));
 			filter5_weights[i][j] = Color(w,w,w,0);
-			sum += 3 * w;
+			sum += w;
 		}
 	}
-	sum /= 3;
+	invSum = 1.0 / sum;
 	for (int i=0; i<5; i++) {
 		for (int j=0; j<5; j++) {
-			for (int k=0; k<3; k++) {
-				filter5_weights[i][j][k] /= sum;
-			}
+			filter5_weights[i][j] *= invSum;
 		}
 	}
 
@@ -199,9 +195,10 @@ int _tmain(int argc, _TCHAR* argv[])
 			for (int x=0; x<width; x++) {
 				const uint8_t idx = quantized_image[y][x];
 				Color col = palette[idx];
-				c[2] = (unsigned char)(255*col[0]);
-				c[1] = (unsigned char)(255*col[1]);
-				c[0] = (unsigned char)(255*col[2]);
+				col *= 255.0;
+				c[2] = (unsigned char)(col[0]);
+				c[1] = (unsigned char)(col[1]);
+				c[0] = (unsigned char)(col[2]);
 				fwrite(c, 3, 1, out);
 			}
 		}

--- a/quantize.cpp
+++ b/quantize.cpp
@@ -74,9 +74,9 @@ void random_permutation(
 	vector<int>& result
 	)
 {
-	result.clear();
+	result.resize(count);
 	for (size_t i=0; i<count; ++i) {
-		result.push_back(i);
+		result[i] = i;
 	}
 	random_shuffle(result.begin(), result.end());
 }
@@ -89,10 +89,11 @@ void random_permutation_2d(
 {
 	vector<int> perm1d;
 	random_permutation(width*height, perm1d);
-	while (!perm1d.empty()) {
-		int idx = perm1d.back();
-		perm1d.pop_back();
-		result.push_back(pair<int,int>(idx % width, idx / width));
+	const size_t sz = perm1d.size();
+	result.resize(sz);
+	for (size_t i=0; i<sz; ++i) {
+		int idx = perm1d[sz-1-i];
+		result[i] = pair<int,int>(idx % width, idx / width);
 	}
 }
 
@@ -291,17 +292,11 @@ void compute_initial_s(
 	const Image& b
 	)
 {
-	init_image(s);
 	size_t palette_size  = s.width_;
 	int coarse_width  = coarse_variables.width_;
 	int coarse_height = coarse_variables.height_;
 	int center_x = (b.width_-1)/2, center_y = (b.height_-1)/2;
-
-	printf(
-		"%s started %d %d %d\n",
-		__FUNCTION__, palette_size, coarse_width, coarse_height
-	);
-
+	
 	Color center_b = b_value(b,0,0,0,0);
 	Color zero_vector;
 	zero_vector.zero();
@@ -519,10 +514,12 @@ void spatial_color_quant(
 	size_t iters_at_current_level = 0;
 	bool skip_palette_maintenance = false;
 	Image s(num_colors, num_colors);
+	init_image(s);
 	compute_initial_s(s, coarse_variables, *b_vec[coarse_level]);
 	Image* j_palette_sum =
 		new Image(coarse_variables.width_, coarse_variables.height_);
 	compute_initial_j_palette_sum(*j_palette_sum, coarse_variables, palette, num_colors);
+	vector<double> meanfield_logs(num_colors), meanfields(num_colors);
 	while (coarse_level >= 0 || temperature > final_temperature) {
 		// Need to reseat this reference in case we changed p_coarse_variables
 		Array3D<double>& coarse_variables = *p_coarse_variables;
@@ -545,7 +542,6 @@ void spatial_color_quant(
 			while (!visit_queue.empty()) {
 				// If we get to 10% above initial size, just revisit them all
 				if (visit_queue.size() > coarse_variables.width_*coarse_variables.height_*11.0/10) {
-					visit_queue.clear();
 					random_permutation_2d(coarse_variables.width_, coarse_variables.height_, visit_queue);
 				}
 
@@ -557,11 +553,12 @@ void spatial_color_quant(
 				Color p_i;
 				p_i.zero();
 				for (int y=0; y<b.height_; ++y) {
+					int j_y = y - center_y + i_y;
+					if (j_y < 0 || j_y >= coarse_variables.height_) continue;
 					for (int x=0; x<b.width_; ++x) {
 						int j_x = x - center_x + i_x;
-						int j_y = y - center_y + i_y;
 						if (i_x == j_x && i_y == j_y) continue;
-						if (j_x < 0 || j_y < 0 || j_x >= coarse_variables.width_ || j_y >= coarse_variables.height_) continue;
+						if (j_x < 0 || j_x >= coarse_variables.width_) continue;
 						Color b_ij = b_value(b, i_x, i_y, j_x, j_y);
 						Color j_pal = (*j_palette_sum)[j_y][j_x];
 						p_i += b_ij * j_pal;
@@ -570,7 +567,6 @@ void spatial_color_quant(
 				p_i *= 2.0;
 				p_i += a[i_y][i_x];
 
-				vector<double> meanfield_logs, meanfields;
 				double max_meanfield_log = -numeric_limits<double>::infinity();
 				double meanfield_sum = 0.0;
 				for (size_t v=0; v<num_colors; ++v) {
@@ -580,14 +576,15 @@ void spatial_color_quant(
 					// will choose a value that makes the maximum e^100.
 					Color p_i2; p_i2 = p_i;
 					double m = -(palette[v].dot_product(p_i2 + middle_b.direct_product(palette[v]))) / temperature;
-					meanfield_logs.push_back(m);
+					meanfield_logs[v] = m;
 					if (m > max_meanfield_log) {
 						max_meanfield_log = m;
 					}
 				}
 				for (size_t v=0; v<num_colors; ++v) {
-					meanfields.push_back(exp(meanfield_logs[v]-max_meanfield_log+100));
-					meanfield_sum += meanfields.back();
+					double d = exp(meanfield_logs[v]-max_meanfield_log+100);
+					meanfields[v] = d;
+					meanfield_sum += d;
 				}
 				if (meanfield_sum == 0) {
 					cout << "Fatal error: Meanfield sum underflowed. Please contact developer." << endl;

--- a/stdafx.h
+++ b/stdafx.h
@@ -10,6 +10,7 @@
 #include <stdio.h>
 #include <tchar.h>
 
+#define _SECURE_SCL 0
 
 
 // TODO: プログラムに必要な追加ヘッダーをここで参照してください。