Cross-Platform (Win32, Linux) upto Hardware Texturing (scanline order, nearest sampling only)

.gitignore

@@ -7,8 +7,9 @@
 # objs
 datasets
 
-# build
-build
+# directories
+[Bb]uild*
+venv
 
 ## log info
 *_log.txt

include/rtm/aabb.hpp

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <cfloat>
 #include "float.hpp"
 #include "int.hpp"
 #include "vec3.hpp"

include/rtm/bvh.hpp

@@ -1,5 +1,6 @@
 #pragma once
 
+#include "macros.hpp"
 #include "int.hpp"
 #include "aabb.hpp"
 #include "mesh.hpp"
@@ -718,7 +719,7 @@ class BVH
 					cache.decisions[j] = cache.decisions[j - 1];
 			}
 
-			if(cache.decisions[1].cost == INFINITY) __debugbreak();
+			if(cache.decisions[1].cost == INFINITY) add_breakpoint();
 		}
 
 		struct CostItem

include/rtm/float.hpp

@@ -5,10 +5,20 @@
 
 #ifndef __riscv
 #include <cmath>
-#include <intrin.h>
+	#if defined BUILD_PLATFORM_WINDOWS
+		#include <intrin.h>
+	#elif defined BUILD_PLATFORM_LINUX
+		#include <immintrin.h>
+	#endif
 #include <cfenv>
 #endif
 
+#ifdef BUILD_PLATFORM_LINUX
+	// Declarations of intrinsics used that are defined but not declared in immintrin
+	extern __m128 _mm_cos_ps(__m128 __A);
+	extern __m128 _mm_sin_ps(__m128 __A);
+#endif
+
 namespace rtm
 {
 
@@ -157,7 +167,7 @@ inline int32_t f32_to_i24(float f32, uint8_t max_exp = 127, int rounding = 0)
 	else if(rounding == 1)
 		norm = std::ceil(norm);
 
-	if(norm > ((1 << 23) - 1) || norm < -(1 << 23)) __debugbreak();
+	if(norm > ((1 << 23) - 1) || norm < -(1 << 23)) add_breakpoint();
 
 	return (int32_t)norm;
 }
@@ -185,7 +195,7 @@ inline uint16_t f32_to_i16(float f32, uint8_t max_exp = 127, int rounding = 0)
 	else if(rounding == 1)
 		norm = std::ceil(norm);
 
-	if(norm > ((1 << 15) - 1) || norm < -(1 << 15)) __debugbreak();
+	if(norm > ((1 << 15) - 1) || norm < -(1 << 15)) add_breakpoint();
 	return (int16_t)norm;
 }
 #endif

include/rtm/macros.hpp

@@ -2,4 +2,18 @@
 
 #if defined __x86_64__ || defined _M_X64
 #define __x86
+#endif
+
+#if defined _WIN16 || defined WIN32 || defined _WIN32 || defined WIN64 || defined _WIN64 || defined __WIN32__ || defined __TOS_WIN__ || defined __WINDOWS__
+	#define BUILD_PLATFORM_WINDOWS
+#elif defined __linux__
+	#define BUILD_PLATFORM_LINUX
+#endif
+
+//To add breakpoints for debugging at runtime
+#if defined BUILD_PLATFORM_WINDOWS
+	#define add_breakpoint() __debugbreak()
+#elif defined BUILD_PLATFORM_LINUX
+	#include <signal.h>
+	#define add_breakpoint() raise(SIGINT)
 #endif

include/rtm/mesh.hpp

@@ -67,8 +67,14 @@ class Mesh
 public:
 	Mesh(std::string file_path) : mtl_lib("")
 	{
- 		load_obj(file_path.c_str());
-		load_mtl(swap_path(mtl_lib, "../../../datasets/"), "../../../datasets/textures/");
+ 		if(!load_obj(file_path.c_str())) {
+			printf("Mesh: failed to load mesh OBJ file %s\n", file_path.c_str());
+			return;
+		}
+		if(!load_mtl(swap_path(mtl_lib, "datasets/"), "datasets/textures/")) {
+			printf("Mesh: failed to load mesh textures or MTL file %s\n", mtl_lib);
+			return;
+		}
 	}
 
 	static inline rtm::vec2 read_vec2(char* line)

include/rtm/texture.hpp

@@ -21,8 +21,13 @@ class Texture2D
 	int32_t comp;
 	Texel* texels;
 
+private:
+	// allows Arches to shallow copy textures
+	// not to be used by kernels!!!
+	int32_t *pRefCount;
+
 public:
-	Texture2D() : texels(nullptr) {};
+	Texture2D() : texels(nullptr), pRefCount(nullptr) {};
 #ifndef __riscv
 	Texture2D(std::string filename)
 	{
@@ -39,80 +44,108 @@ class Texture2D
 					texels[i].channel[j] = data[i * comp + j];
 
 			stbi_image_free(data);
+
+			pRefCount = (int32_t *)malloc(sizeof(int32_t));
+			*pRefCount = 1;  // acquired first time by direct assignment
+
 			printf("Loaded: %s \n", filename.c_str());
 		}
 		else
 		{
 			texels = nullptr;
+			pRefCount = nullptr;
 			printf("Failed: %s \n", filename.c_str());
 		}
 	}
 
-	Texture2D(const Texture2D& other)
+	Texture2D(const Texture2D& other) : width(other.width), height(other.height), comp(other.comp)
+    {
+		other.acquire();
+		texels = other.texels;
+		pRefCount = other.pRefCount;
+    }
+
+	Texture2D& operator=(const Texture2D& other) 
+    {
+        if (this == &other)
+			return *this;
+
+		// copy other members
+		width = other.width;
+        height = other.height;
+        comp = other.comp;
+
+		// following order is safer than 'first release, then acquire'
+		other.acquire();       // prebook ownership of new resource
+		release();             // release ownership of current resource
+		texels = other.texels; // acquire ownership of new resource
+		pRefCount = other.pRefCount;
+        return *this;
+    }
+
+	~Texture2D()
 	{
-		memcpy(this, &other, sizeof(Texture2D));
-		uint32_t size = sizeof(Texel) * width * height;
-		texels = (Texel*)malloc(size);
-		memcpy(texels, other.texels, size);
+		release();
 	}
+#endif
 
-	Texture2D& operator=(const Texture2D& other)
+	rtm::vec4 sample(const rtm::vec2& uv) const
 	{
-		if(texels) free(texels);
-		memcpy(this, &other, sizeof(Texture2D));
-		uint32_t size = sizeof(Texel) * width * height;
-		texels = (Texel*)malloc(size);
-		memcpy(texels, other.texels, size);
-		return *this;
+		if(width == 0 && height == 0) return rtm::vec4(0.0f);
+		return read_nearest(uv);
 	}
 
-	~Texture2D()
+	rtm::uvec2 get_iuv(const rtm::vec2& uv, const rtm::vec2& offset = rtm::vec2(0.0f)) const
 	{
-		if(texels) 
-			free(texels);
+		rtm::vec2 fuv = uv * rtm::vec2(width, height) + offset;
+		return rtm::uvec2(fuv[0], fuv[1]);
 	}
-#endif
 
-	rtm::vec3 sample(const rtm::vec2& uv) const
+	Texel* get_texel_addr(const rtm::uvec2& iuv) const
 	{
-		if(width == 0 && height == 0) return rtm::vec3(0.0f);
-		if(width == 1 && height == 1) return read({0, 0});
-
-		rtm::vec2 _uv = rtm::mod(uv, rtm::vec2(1.0f)) * rtm::vec2(width, height);
-		//return read_nearest(_uv);
-
-		//rtm::vec2 _uv = rtm::mod(uv, rtm::vec2(1.0f)) * rtm::vec2(width, height);
-		rtm::vec3 s00 = read_nearest(_uv + rtm::vec2(-0.5f, -0.5f));
-		rtm::vec3 s10 = read_nearest(_uv + rtm::vec2(0.5f, -0.5f));
-		rtm::vec3 s01 = read_nearest(_uv + rtm::vec2(-0.5f, 0.5f));
-		rtm::vec3 s11 = read_nearest(_uv + rtm::vec2(0.5f, 0.5f));
-
-		rtm::vec2 ic = rtm::mod(_uv + rtm::vec2(0.5f), rtm::vec2(1.0f));
-		rtm::vec3 s0 = rtm::mix(s00, s01, ic.y);
-		rtm::vec3 s1 = rtm::mix(s10, s11, ic.y);
-
-		return rtm::mix(s0, s1, ic.x);
+		uint32_t x = iuv[0] % width;
+		uint32_t y = iuv[1] % height;
+		return &texels[y * width + x];
 	}
 
-private:
-	rtm::vec3 read(const rtm::uvec2& iuv) const
+	static rtm::vec4 decode_texel(const Texel& texel)
 	{
-		uint i = iuv[1] * width + iuv[0];
-		return rtm::vec3(texels[i].channel[0], texels[i].channel[1], texels[i].channel[2]) * (1.0f / 255.0f);
+		return rtm::vec4(texel.channel[0], texel.channel[1], texel.channel[2], texel.channel[3]) * (1.0f / 255.0f);
 	}
 
-	rtm::uvec2 get_nearest(const rtm::vec2& uv) const
+	rtm::vec4 read_nearest(const rtm::vec2& uv, rtm::vec2 offset = rtm::vec2(0.0f)) const
 	{
-		rtm::uvec2 mv = rtm::uvec2(width, height);
-		rtm::uvec2 iuv = rtm::uvec2(uv[0], uv[1]) + mv;
-		iuv[0] = iuv[0] % mv[0]; iuv[1] = iuv[1] % mv[1];
-		return  iuv;
+		rtm::uvec2 iuv = get_iuv(uv, offset);
+		Texel* texel = get_texel_addr(iuv);
+		return decode_texel(*texel);
 	}
 
-	rtm::vec3 read_nearest(const rtm::vec2& uv) const
+#ifndef __riscv
+private:
+	void acquire(void) const
 	{
-		rtm::vec3 v = read(get_nearest(uv));
-		return v;
+		if (!pRefCount)
+			return;
+
+		(*pRefCount)++;
 	}
+
+	void release(void)
+	{
+		if (!pRefCount)
+			return;
+
+		(*pRefCount)--;
+		if((*pRefCount) <= 0)
+		{
+			if(texels)
+				free(texels);
+			texels = nullptr;
+
+			free(pRefCount);
+			pRefCount = nullptr;
+		}
+	}
+#endif
 };
 

scripts/log-parser.py

@@ -39,6 +39,10 @@ def gauss(n=11,sigma=1):
                 data[i].append(float(re.findall(r'[\d]*[.][\d]+', line)[0]))
 
 
+print(f"Total Cycles found: {len(cycles)}")
+for i in range(len(keys)):
+    print(f"Key '{keys[i]}' found {len(data[i])} times")
+
 show_unsmoothed = True
 if show_unsmoothed:
     for i in range(len(keys)):
@@ -49,13 +53,33 @@ def gauss(n=11,sigma=1):
 kernel = np.array(gauss(kernel_size, 5))
 kernel = kernel / kernel.sum()
 
+# for i in range(len(keys)):
+#     if len(data[i]) == len(cycles):
+#         plt.plot(cycles, np.convolve(data[i], kernel, mode='same'), colors[i], label=keys[i])
+
 for i in range(len(keys)):
-    if len(data[i]) == len(cycles):
-        plt.plot(cycles, np.convolve(data[i], kernel, mode='same'), colors[i], label=keys[i])
+    # determine the shortest length to keep (x, y) in sync
+    min_len = min(len(data[i]), len(cycles))
+
+    if min_len > 0:
+        # slice both to min_len so (x, y) match perfectly
+        x = cycles[:min_len]
+        y = data[i][:min_len]
+
+        # plot unsmoothed (faded)
+        plt.plot(x, y, colors[i], alpha=0.125)
+
+        # plot smoothed (solid with label)
+        smoothed = np.convolve(y, kernel, mode='same')
+        plt.plot(x, smoothed, colors[i], label=keys[i])
+    else:
+        print(f"Skipping {keys[i]} - no data matches found.")
 
 plt.legend(loc="upper right")
 plt.ylabel(ylabel)
 plt.xlabel('Time (cycles)')
 plt.yscale('log')
 plt.grid()
-plt.show()
+plt.savefig("simulation_results.png") # save instead of show
+print("Plot saved to simulation_results.png")
+#plt.show()

src/arches-v2/CMakeLists.txt

@@ -3,6 +3,11 @@ add_compile_definitions(UNICODE _UNICODE)
 
 set(PROJECT_NAME "arches-v2")
 
+if(LINUX)
+  # enable intrinsics like _lzcnt_u64
+  add_compile_options(-march=native)
+endif()
+
 file(GLOB_RECURSE ALL_INCLUDE CONFIGURE_DEPENDS "*.hpp" "*.h")
 file(GLOB_RECURSE ALL_SOURCES CONFIGURE_DEPENDS "*.cpp" "*.cc")
 file(GLOB_RECURSE USIMM_CONFIG CONFIGURE_DEPENDS "*.vi" "*.cfg")
@@ -53,8 +58,12 @@ target_include_directories(${PROJECT_NAME} PUBLIC ${PROJECT_SOURCE_DIR}/src)
 set_property(DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} PROPERTY VS_STARTUP_PROJECT DISABLE)
 
 set_target_properties(${PROJECT_NAME} PROPERTIES OUTPUT_NAME ${PROJECT_NAME})
-target_link_directories(${PROJECT_NAME} PUBLIC ${PROJECT_SOURCE_DIR}/libraries/tbb)
-target_link_libraries(${PROJECT_NAME} PRIVATE tbb12.lib)
+if(WINDOWS)
+  target_link_directories(${PROJECT_NAME} PUBLIC ${PROJECT_SOURCE_DIR}/libraries/tbb)
+  target_link_libraries(${PROJECT_NAME} PRIVATE tbb12.lib)
+elseif(LINUX)
+  target_link_libraries(${PROJECT_NAME} PRIVATE tbb)
+endif()
 target_link_libraries(${PROJECT_NAME} PRIVATE Ramulator)
 #set_target_properties(${PROJECT_NAME} PROPERTIES RUNTIME_OUTPUT_DIRECTORY_DEBUG ${CMAKE_CURRENT_BINARY_DIR})
 #set_target_properties(${PROJECT_NAME} PROPERTIES RUNTIME_OUTPUT_DIRECTORY_RELEASE ${CMAKE_CURRENT_BINARY_DIR})