posts-article minor changes

Feel like opting for u1-4.. (b1-4 not meaning byte anymore, not sure why I bothered, just bools/bits)

.vscode/cozy-and-windy/themes/cozy-and-windy-color-theme.json

@@ -30,7 +30,7 @@
 		"editorHoverWidget.background": "#2c334b"
 	},
 	"semanticTokenColors": {
-		"comment": { "foreground": "#868686", "fontStyle": "italic" },
+		"comment": { "foreground": "#868686", }, //"fontStyle": "italic" },
 		"keyword": { "foreground": "#d8bd5b" },
 		"string": { "foreground": "#d46a54" },
 		"number": { "foreground": "#b5cea8" },
@@ -81,7 +81,7 @@
 		"tapeDelaySlot": { "foreground": "#ff5647" }
 	},
 	"tokenColors": [
-		{ "scope": ["comment", "comment.block", "comment.line", "comment.block.documentation"], "settings": { "foreground": "#868686", "fontStyle": "italic" } },
+		{ "scope": ["comment", "comment.block", "comment.line", "comment.block.documentation"], "settings": { "foreground": "#868686", } }, //"fontStyle": "italic" } },
 		{ "scope": ["keyword", "keyword.control", "keyword.other"], "settings": { "foreground": "#d8bd5b" } },
 		{ "scope": ["string", "string.quoted"], "settings": { "foreground": "#d46a54" } },
 		{ "scope": ["string.quoted.other"], "settings": { "foreground": "#d69d85" } },

.vscode/tape-atom-syntax/classifier.js

@@ -21,72 +21,108 @@ const TOKEN_TYPES = [
 	"tapeGprRegister",
 	"tapeCop2Register",
 	"tapeDuffleType",
-	"tapeAttribute",
+	"tapeAt__ibute",
 	"keyword",
 	"macro",
 ];
 
-const TOKEN_MODIFIERS = ["declaration", "tapeRead", "tapeWrite", "tapeAuto"];
-const TOKEN_TYPE_INDEX = new Map(TOKEN_TYPES.map((name, index) => [name, index]));
+const TOKEN_MODIFIERS      = ["declaration", "tapeRead", "tapeWrite", "tapeAuto"];
+const TOKEN_TYPE_INDEX     = new Map(TOKEN_TYPES.map((name, index) => [name, index]));
 const TOKEN_MODIFIER_INDEX = new Map(TOKEN_MODIFIERS.map((name, index) => [name, index]));
 
 const ATOM_KEYWORDS = new Set(["MipsAtom_", "MipsAtom_Proc_"]);
 const COMPONENT_KEYWORDS = new Set(["MipsAtomComp_", "MipsAtomComp_Proc_"]);
 const ANNOTATIONS = new Set([
-	"atom_info", "atom_bind", "atom_reads", "atom_writes", "atom_label",
-	"atom_offset", "atom_reg", "atom_type", "atom_ctx", "atom_phase",
-	"atom_auto_reg", "phase_auto_reg", "atom_dbg_skip",
+	"atom_info",
+	"atom_bind",
+	"atom_reads",
+	"atom_writes",
+	"atom_label",
+	"atom_offset",
+	"atom_reg",
+	"atom_type",
+	"atom_ctx",
+	"atom_phase",
+	"atom_auto_reg", 
+	"phase_auto_reg", 
+	"atom_dbg_skip",
 ]);
 
 const DSL_KEYWORDS = new Set([
-	"FI_", "I_", "NI_", "Relative_", "Struct_", "Enum_", "Union_", "Array_",
-	"Slice_", "TypeR_", "TypeV_", "align_", "internal", "local_persist", "global",
-	"RO_", "LP_", "gknown", "expect_", "cexpr_",
+	"enum", "struct", "union",
+	"offset_of", "static_assert", "typeof", "typeof_ptr", "typeof_same",
+	"glue", "tmpl",
+	"A_", "FI_", "I_", "NI_", 
+	"Array_", "Enum_", "Proc_", "Relative_", "Struct_", "Union_", "Slice_", 
+
+	// "TypeR_", "TypeV_",
+
+	"align_", 
+	"internal", "local_persist", "global",
+
+	"RO_", "LP_", 
+	"gknown", "expect_", "cexpr_",
+
+	"O_", "OA_", "S_", "C_", "T_", "T_same", "R_", "V_",
+	"r_", "v_", "rt_", "vt_",
+
 	"asm", "asm_words", "asm_rpins", "asm_clobber",
-	"O_", "S_", "C_", "T_", "tmpl", "glue", "r_", "v_", "tr_", "tv_",
 	"rgcc", "r_use", "r_set", "r_mod", "r_imm", "r_mem",
-	"u1_", "u2_", "u4_", "u8_", "s1_", "s2_", "s4_", "s8_",
-	"u1_r", "u2_r", "u4_r", "u8_r", "u1_v", "u2_v", "u4_v", "u8_v",
+	
+	"b1_", "b2_", "b4_", "b8_", 
+	"u1_", "u2_", "u4_", "u8_", 
+	"s1_", "s2_", "s4_", "s8_",
+	"b1_r", "b2_r", "b4_r", "b8_r", 
+	"b1_v", "b2_v", "b4_v", "b8_v", 
+	"u1_r", "u2_r", "u4_r", "u8_r", 
+	"u1_v", "u2_v", "u4_v", "u8_v",
+
+	"u4_lo", "u4_hi",
 ]);
 
-const DELAY_SLOT_KEYWORDS = new Set(["LdSlot_", "BdSlot_", "DmaSlot_", "GteDelay_"]);
+const DELAY_SLOT_KEYWORDS = new Set([
+	"LdSlot_",
+	"BdSlot_",
+	"DmaSlot_",
+	"GteDelay_"
+]);
 
 const CONTROL_FLOW_PREFIXES = /^(?:branch_|jump_|call_)/;
 
 const ROLE_TO_TYPE = {
-	atomName: "tapeAtomName",
+	atomName:      "tapeAtomName",
 	componentName: "tapeComponentName",
-	bindType: "tapeBindType",
-	duffleType: "tapeDuffleType",
-	gprRegister: "tapeGprRegister",
-	cop2Register: "tapeCop2Register",
+	bindType:      "tapeBindType",
+	duffleType:    "tapeDuffleType",
+	gprRegister:   "tapeGprRegister",
+	cop2Register:  "tapeCop2Register",
 };
 
 function registerType(name, index) {
 	const kind = index.registers.get(name);
-	if (kind === "gpr" || /^R_[A-Za-z0-9_]+$/.test(name)) return "tapeGprRegister";
-	if (kind === "cop2" || /^(?:C2_|gte_cr_)[A-Za-z0-9_]+$/.test(name)) return "tapeCop2Register";
+	if (kind === "gpr"  || /^R_[A-Za-z0-9_]              + $/.test(name)) return "tapeGprRegister";
+	if (kind === "cop2" || /^(?:C2_|gte_cr_)[A-Za-z0-9_] + $/.test(name)) return "tapeCop2Register";
 	return null;
 }
 
 function instructionType(name, index) {
 	const domain = index.macros.get(name);
 	if (domain === "control") return "tapeControlFlow";
-	if (domain === "cpu") return "tapeCpuInstruction";
-	if (domain === "gte") return "tapeGteInstruction";
-	if (domain === "gpu") return "tapeGpuInstruction";
+	if (domain === "cpu")     return "tapeCpuInstruction";
+	if (domain === "gte")     return "tapeGteInstruction";
+	if (domain === "gpu")     return "tapeGpuInstruction";
 	if (domain === "component") {
 		if (/^mac_gte_/.test(name)) return "tapeGteInstruction";
-		if (/^mac_gp/.test(name)) return "tapeGpuInstruction";
-		if (/^mac_/.test(name)) return "tapeComponentInstruction";
+		if (/^mac_gp/.test(name))   return "tapeGpuInstruction";
+		if (/^mac_/.test(name))     return "tapeComponentInstruction";
 		return "macro";
 	}
-	if (domain === "utility") return "macro";
+	if (domain === "utility")      return "macro";
 	if (/^gte_(?!cr_)/.test(name)) return "tapeGteInstruction";
-	if (/^gp[01]_/.test(name)) return "tapeGpuInstruction";
-	if (/^mac_gte_/.test(name)) return "tapeGteInstruction";
-	if (/^mac_gp/.test(name)) return "tapeGpuInstruction";
-	if (/^mac_/.test(name)) return "tapeComponentInstruction";
+	if (/^gp[01]_/.test(name))     return "tapeGpuInstruction";
+	if (/^mac_gte_/.test(name))    return "tapeGteInstruction";
+	if (/^mac_gp/.test(name))      return "tapeGpuInstruction";
+	if (/^mac_/.test(name))        return "tapeComponentInstruction";
 	return null;
 }
 
@@ -100,13 +136,11 @@ function modifierMask(modifiers) {
 }
 
 function isRegUseAccess(tokens, tokenIndex) {
-	const prev = tokens[tokenIndex - 1];
-	if (!prev || prev.text !== ".") return false;
-	const prevPrev = tokens[tokenIndex - 2];
-	if (!prevPrev || prevPrev.kind !== "identifier") return false;
-	const next = tokens[tokenIndex + 1];
+	const prev     = tokens[tokenIndex - 1]; if (!prev || prev.text !== ".") return false;
+	const prevPrev = tokens[tokenIndex - 2]; if (!prevPrev || prevPrev.kind !== "identifier") return false;
+	const next     = tokens[tokenIndex + 1];
 	if (next && next.text === ".") return false;
-	if (prevPrev.text === "r") return true;
+	if (prevPrev.text     === "r") return true;
 	const prev3 = tokens[tokenIndex - 3];
 	const prev4 = tokens[tokenIndex - 4];
 	if (prev3 && prev3.text === "." && prev4 && prev4.kind === "identifier" && prev4.text === "r") return true;
@@ -115,81 +149,92 @@ function isRegUseAccess(tokens, tokenIndex) {
 
 function classifyDocument(source, filePath, workspaceIndex, shouldCancel = () => false) {
 	const scanned = scanSource(source, filePath);
-	const index = mergeIndexes(workspaceIndex, scanned.index);
-	const spans = [];
+	const index   = mergeIndexes(workspaceIndex, scanned.index);
+	const spans   = [];
 
 	for (let tokenIndex = 0; tokenIndex < scanned.tokens.length; tokenIndex += 1) {
 		if (shouldCancel()) break;
-		const token = scanned.tokens[tokenIndex];
-		if (token.kind !== "identifier") continue;
+		const token = scanned.tokens[tokenIndex]; if (token.kind !== "identifier") continue;
 
-		let type = null;
-		let modifiers = [];
+		let type          = null;
+		let modifiers     = [];
 		const declaration = scanned.declarations.get(token.start);
-		const context = nearestCall(scanned.contexts, tokenIndex);
+		const context     = nearestCall(scanned.contexts, tokenIndex);
 
 		if (declaration) {
-			type = ROLE_TO_TYPE[declaration.role] || null;
+			type      = ROLE_TO_TYPE[declaration.role] || null;
 			modifiers = declaration.modifiers.slice();
-		} else if (ATOM_KEYWORDS.has(token.text)) {
+		} 
+		else if (ATOM_KEYWORDS.has(token.text)) {
 			type = "tapeAtomKeyword";
-		} else if (COMPONENT_KEYWORDS.has(token.text)) {
+		} 
+		else if (COMPONENT_KEYWORDS.has(token.text)) {
 			type = "keyword";
-		} else if (ANNOTATIONS.has(token.text)) {
+		} 
+		else if (ANNOTATIONS.has(token.text)) {
 			type = "tapeAnnotation";
-		} else if (context && context.callee === "atom_bind" && context.argIndex === 0) {
+		} 
+		else if (context && context.callee === "atom_bind" && context.argIndex === 0) {
 			type = "tapeBindType";
-		} else if (context && context.callee === "atom_phase" && context.argIndex === 0) {
+		} 
+		else if (context && context.callee === "atom_phase" && context.argIndex === 0) {
 			type = "tapePhase";
 			modifiers = ["declaration"];
-		} else if (context && context.callee === "atom_ctx" && context.argIndex === 0) {
+		} 
+		else if (context && context.callee === "atom_ctx" && context.argIndex === 0) {
 			type = "tapeAtomName";
-		} else if (context && context.callee === "atom_label" && context.argIndex === 0) {
+		} 
+		else if (context && context.callee === "atom_label" && context.argIndex === 0) {
 			type = "tapeLabel";
 			modifiers = ["declaration"];
-		} else if (context && context.callee === "atom_offset" && context.argIndex <= 1) {
+		} 
+		else if (context && context.callee === "atom_offset" && context.argIndex <= 1) {
 			type = "tapeLabel";
-		} else if (context && context.callee === "atom_reads") {
+		} 
+		else if (context && context.callee === "atom_reads") {
 			type = registerType(token.text, index);
 			if (type) modifiers = ["tapeRead"];
-		} else if (context && context.callee === "atom_writes") {
+		} 
+		else if (context && context.callee === "atom_writes") {
 			type = registerType(token.text, index);
 			if (type) modifiers = ["tapeWrite"];
-		} else if (context && context.callee === "atom_auto_reg") {
+		} 
+		else if (context && context.callee === "atom_auto_reg") {
 			if (context.argIndex === 0) type = "tapeAtomName";
 			if (context.argIndex === 1) {
-				type = "tapeGprRegister";
+				type      = "tapeGprRegister";
 				modifiers = ["declaration", "tapeAuto"];
 			}
-		} else if (context && context.callee === "phase_auto_reg") {
+		} 
+		else if (context && context.callee === "phase_auto_reg") {
 			if (context.argIndex === 0) type = "tapePhase";
 			if (context.argIndex === 1) {
-				type = "tapeGprRegister";
+				type      = "tapeGprRegister";
 				modifiers = ["declaration", "tapeAuto"];
 			}
 		}
 
-		if (!type && index.bindTypes.has(token.text)) type = "tapeBindType";
-		if (!type && DSL_KEYWORDS.has(token.text)) type = "keyword";
-		if (!type && index.types.has(token.text)) type = "tapeDuffleType";
-		if (!type && index.attributes.has(token.text)) type = "tapeAttribute";
-		if (!type) type = registerType(token.text, index);
-		if (!type && DELAY_SLOT_KEYWORDS.has(token.text)) type = "tapeDelaySlot";
-		if (!type) {
+		if (! type && index.bindTypes.has(token.text))     type = "tapeBindType";
+		if (! type && DSL_KEYWORDS.has(token.text))        type = "keyword";
+		if (! type && index.types.has(token.text))         type = "tapeDuffleType";
+		if (! type && index.attributes.has(token.text))    type = "tapeAttribute";
+		if (! type)                                        type = registerType(token.text, index);
+		if (! type && DELAY_SLOT_KEYWORDS.has(token.text)) type = "tapeDelaySlot";
+		if (! type) {
 			const domain = index.macros.get(token.text);
 			if (domain === "control" || (domain && CONTROL_FLOW_PREFIXES.test(token.text))) {
 				type = "tapeControlFlow";
 			}
 		}
-		if (!type && isRegUseAccess(scanned.tokens, tokenIndex)) type = "tapeGprRegister";
-		if (!type) type = instructionType(token.text, index);
-		if (!type && /^(?:Slice_|A[0-9]+_)/.test(token.text)) type = "tapeDuffleType";
-		if (!type && /_[RV]$/.test(token.text)) type = "tapeDuffleType";
-		if (!type && index.atoms.has(token.text)) type = "tapeAtomName";
-		if (!type && index.components.has(token.text)) type = "tapeComponentName";
-		if (!type && index.phases.has(token.text)) type = "tapePhase";
-		if (!type && index.labels.has(token.text)) type = "tapeLabel";
-		if (!type) continue;
+		if (! type && isRegUseAccess(scanned.tokens, tokenIndex)) type = "tapeGprRegister";
+		if (! type)                                               type = instructionType(token.text, index);
+		if (! type && /^(?:Slice_|A[0-9]+_)/.test(token.text))    type = "tapeDuffleType";
+		if (! type && /_[RV]$/.test(token.text))                  type = "tapeDuffleType";
+		if (! type && index.atoms.has(token.text))                type = "tapeAtomName";
+		if (! type && index.components.has(token.text))           type = "tapeComponentName";
+		if (! type && index.phases.has(token.text))               type = "tapePhase";
+		if (! type && index.labels.has(token.text))               type = "tapeLabel";
+		if (! type) continue;
 
 		spans.push({
 			text: token.text,

.vscode/tape-atom-syntax/extension.js

@@ -2,10 +2,10 @@
 
 const vscode = require("vscode");
 const { TOKEN_MODIFIERS, TOKEN_TYPES, classifyDocument } = require("./classifier");
-const { createIndex, mergeIndexes, scanSource } = require("./source-index");
+const { createIndex, mergeIndexes, scanSource }          = require("./source-index");
 
-const SOURCE_GLOB = "**/*.{c,h,cc,cpp,cxx,hh,hpp,hxx}";
-const EXCLUDE_GLOB = "**/{gen,build,.slop_cache,toolchain,node_modules}/**";
+const SOURCE_GLOB        = "**/*.{c,h,cc,cpp,cxx,hh,hpp,hxx}";
+const EXCLUDE_GLOB       = "**/{gen,build,.slop_cache,toolchain,node_modules}/**";
 const EXCLUDED_SEGMENTS = new Set(["gen", "build", ".slop_cache", "toolchain", "node_modules"]);
 
 function isExcluded(uri) {
@@ -18,28 +18,29 @@ function formatError(filePath, error) {
 }
 
 async function activate(context) {
-	const output = vscode.window.createOutputChannel("Tape Atom DSL");
+	const output  = vscode.window.createOutputChannel("Tape Atom DSL");
 	const emitter = new vscode.EventEmitter();
-	const legend = new vscode.SemanticTokensLegend(TOKEN_TYPES, TOKEN_MODIFIERS);
-	let workspaceIndex = createIndex();
+	const legend  = new vscode.SemanticTokensLegend(TOKEN_TYPES, TOKEN_MODIFIERS);
+	let workspaceIndex    = createIndex();
 	let rebuildGeneration = 0;
-	let debounceHandle = null;
+	let debounceHandle    = null;
 
 	async function rebuildIndex() {
 		const generation = ++rebuildGeneration;
-		const files = await vscode.workspace.findFiles(SOURCE_GLOB, EXCLUDE_GLOB);
-		let nextIndex = createIndex();
+		const files      = await vscode.workspace.findFiles(SOURCE_GLOB, EXCLUDE_GLOB);
+		let nextIndex    = createIndex();
 
 		for (const uri of files) {
 			if (generation !== rebuildGeneration) return;
 			if (isExcluded(uri)) continue;
 			try {
-				const bytes = await vscode.workspace.fs.readFile(uri);
+				const bytes  = await vscode.workspace.fs.readFile(uri);
 				const source = Buffer.from(bytes).toString("utf8");
 				const result = scanSource(source, uri.fsPath);
-				nextIndex = mergeIndexes(nextIndex, result.index);
+				nextIndex    = mergeIndexes(nextIndex, result.index);
 				for (const error of result.errors) output.appendLine(formatError(uri.fsPath, error));
-			} catch (error) {
+			} 
+			catch (error) {
 				output.appendLine(`${uri.fsPath}: ${error.stack || error.message || error}`);
 			}
 		}
@@ -53,9 +54,11 @@ async function activate(context) {
 		if (uri && isExcluded(uri)) return;
 		if (debounceHandle !== null) clearTimeout(debounceHandle);
 		debounceHandle = setTimeout(() => {
-			debounceHandle = null;
-			rebuildIndex().catch((error) => output.appendLine(error.stack || String(error)));
-		}, 100);
+				debounceHandle = null;
+				rebuildIndex().catch((error) => output.appendLine(error.stack || String(error)));
+			}, 
+			100
+		);
 	}
 
 	const provider = {
@@ -77,7 +80,8 @@ async function activate(context) {
 					output.appendLine(formatError(document.uri.fsPath || document.uri.toString(), error));
 				}
 				return builder.build();
-			} catch (error) {
+			}
+			catch (error) {
 				output.appendLine(`${document.uri}: ${error.stack || error.message || error}`);
 				return new vscode.SemanticTokensBuilder(legend).build();
 			}
@@ -85,8 +89,8 @@ async function activate(context) {
 	};
 
 	const selector = [
-		{ language: "c", scheme: "file" },
-		{ language: "c", scheme: "untitled" },
+		{ language: "c",   scheme: "file" },
+		{ language: "c",   scheme: "untitled" },
 		{ language: "cpp", scheme: "file" },
 		{ language: "cpp", scheme: "untitled" },
 	];

.vscode/tape-atom-syntax/lexer.js

@@ -10,29 +10,30 @@ function isIdentifierContinue(code) {
 	return isIdentifierStart(code) || (code >= 48 && code <= 57);
 }
 
-function lex(source) {
+function lex(source)
+{
 	if (typeof source !== "string") throw new TypeError("source must be a string");
 
-	const tokens = [];
-	const errors = [];
-	let offset = 0;
-	let line = 0;
-	let character = 0;
+	const tokens    = [];
+	const errors    = [];
+	let   offset    = 0;
+	let   line      = 0;
+	let   character = 0;
 
 	function advance() {
 		if (source[offset] === "\r" && source[offset + 1] === "\n") {
-			offset += 2;
-			line += 1;
+			offset   += 2;
+			line     += 1;
 			character = 0;
 			return;
 		}
 		if (source[offset] === "\n") {
-			offset += 1;
-			line += 1;
+			offset   += 1;
+			line     += 1;
 			character = 0;
 			return;
 		}
-		offset += 1;
+		offset    += 1;
 		character += 1;
 	}
 
@@ -47,7 +48,8 @@ function lex(source) {
 		});
 	}
 
-	while (offset < source.length) {
+	while (offset < source.length)
+	{
 		const ch = source[offset];
 
 		if (/\s/.test(ch)) {
@@ -60,12 +62,14 @@ function lex(source) {
 			continue;
 		}
 
-		if (ch === "/" && source[offset + 1] === "*") {
+		if (ch === "/" && source[offset + 1] === "*")
+		{
 			const start = offset;
 			advance();
 			advance();
 			let closed = false;
-			while (offset < source.length) {
+			while (offset < source.length)
+			{
 				if (source[offset] === "*" && source[offset + 1] === "/") {
 					advance();
 					advance();
@@ -78,12 +82,14 @@ function lex(source) {
 			continue;
 		}
 
-		if (ch === "\"" || ch === "'") {
+		if (ch === "\"" || ch === "'")
+		{
 			const quote = ch;
 			const start = offset;
 			advance();
 			let closed = false;
-			while (offset < source.length) {
+			while (offset < source.length)
+			{
 				if (source[offset] === "\\") {
 					advance();
 					if (offset < source.length) advance();
@@ -97,14 +103,14 @@ function lex(source) {
 				if (source[offset] === "\n" || source[offset] === "\r") break;
 				advance();
 			}
-			if (!closed) errors.push({ kind: "unterminated-literal", offset: start });
+			if (! closed) errors.push({ kind: "unterminated-literal", offset: start });
 			continue;
 		}
 
 		const code = source.charCodeAt(offset);
 		if (isIdentifierStart(code)) {
-			const start = offset;
-			const startLine = line;
+			const start          = offset;
+			const startLine      = line;
 			const startCharacter = character;
 			advance();
 			while (offset < source.length && isIdentifierContinue(source.charCodeAt(offset))) advance();
@@ -112,8 +118,8 @@ function lex(source) {
 			continue;
 		}
 
-		const start = offset;
-		const startLine = line;
+		const start          = offset;
+		const startLine      = line;
 		const startCharacter = character;
 		advance();
 		pushToken("punctuation", start, startLine, startCharacter);
@@ -124,9 +130,9 @@ function lex(source) {
 
 function buildCallContexts(tokens) {
 	const contexts = Array.from({ length: tokens.length }, () => []);
-	const calls = [];
+	const calls  = [];
 	const errors = [];
-	const stack = [];
+	const stack  = [];
 
 	for (let tokenIndex = 0; tokenIndex < tokens.length; tokenIndex += 1) {
 		const token = tokens[tokenIndex];
@@ -134,7 +140,8 @@ function buildCallContexts(tokens) {
 		if (token.text === ")") {
 			if (stack.length === 0) {
 				errors.push({ kind: "unmatched-close-paren", offset: token.start });
-			} else {
+			}
+			else {
 				const frame = stack.pop();
 				if (frame.callee !== null) calls.push({ ...frame, closeTokenIndex: tokenIndex });
 			}
@@ -143,20 +150,20 @@ function buildCallContexts(tokens) {
 		contexts[tokenIndex] = stack
 			.filter((frame) => frame.callee !== null)
 			.map((frame) => ({
-				callee: frame.callee,
+				callee:           frame.callee,
 				calleeTokenIndex: frame.calleeTokenIndex,
-				openTokenIndex: frame.openTokenIndex,
-				argIndex: frame.argIndex,
+				openTokenIndex:   frame.openTokenIndex,
+				argIndex:         frame.argIndex,
 			}));
 
 		if (token.text === "(") {
-			const previous = tokens[tokenIndex - 1];
+			const previous  = tokens[tokenIndex - 1];
 			const hasCallee = previous && previous.kind === "identifier";
 			stack.push({
-				callee: hasCallee ? previous.text : null,
+				callee:           hasCallee ? previous.text : null,
 				calleeTokenIndex: hasCallee ? tokenIndex - 1 : -1,
-				openTokenIndex: tokenIndex,
-				argIndex: 0,
+				openTokenIndex:   tokenIndex,
+				argIndex:         0,
 			});
 			continue;
 		}

.vscode/tape-atom-syntax/source-index.js

@@ -4,8 +4,10 @@ const path = require("node:path");
 const { buildCallContexts, lex, nearestCall } = require("./lexer");
 
 const BASE_TYPES = [
-	"B1", "B2", "B4", "B8", "F4", "F8", "S1", "S2", "S4", "S8",
-	"U1", "U2", "U4", "U8", "MipsAtom", "MipsCode", "Reg",
+	"B1", "B2", "B4", "B8",
+	"F4", "F8", "S1", "S2", "S4", "S8",
+	"U1", "U2", "U4", "U8",
+	"MipsAtom", "MipsCode", "Reg",
 ];
 
 const C_BUILTINS = new Set([
@@ -15,11 +17,13 @@ const C_BUILTINS = new Set([
 ]);
 
 const BASE_ATTRIBUTES = [
-	"FI_", "I_", "NI_", "Relative_", "Struct_", "Enum_", "Union_", "Array_",
-	"Slice_", "TypeR_", "TypeV_", "align_", "internal", "local_persist", "global",
-	"RO_", "LP_", "gknown", "expect_", "cexpr_",
+	"FI_", "I_", "NI_", 
+	"Relative_", "Struct_", "Enum_", "Union_", "Array_", "Slice_", 
+	"align_", "internal", "local_persist", "global",
+	"RO_", "LP_", 
+	"gknown", "expect_", "cexpr_",
 	"asm", "asm_words", "asm_rpins", "asm_clobber",
-	"O_", "S_", "C_", "T_", "tmpl", "glue", "r_", "v_", "tr_", "tv_",
+	"O_", "OA_", "S_", "C_", "T_", "tmpl", "glue", "r_", "v_", "rt_", "vt_",
 	"rgcc", "r_use", "r_set", "r_mod", "r_imm", "r_mem",
 	"u1_", "u2_", "u4_", "u8_", "s1_", "s2_", "s4_", "s8_",
 	"u1_r", "u2_r", "u4_r", "u8_r", "u1_v", "u2_v", "u4_v", "u8_v",
@@ -27,16 +31,16 @@ const BASE_ATTRIBUTES = [
 
 function createIndex() {
 	return {
-		atoms: new Set(),
-		components: new Set(),
+		atoms:            new Set(),
+		components:       new Set(),
 		componentAliases: new Set(),
-		macros: new Map(),
-		registers: new Map(),
-		bindTypes: new Set(),
-		types: new Set(BASE_TYPES),
-		phases: new Set(),
-		labels: new Set(),
-		attributes: new Set(BASE_ATTRIBUTES),
+		macros:           new Map(),
+		registers:        new Map(),
+		bindTypes:        new Set(),
+		types:            new Set(BASE_TYPES),
+		phases:           new Set(),
+		labels:           new Set(),
+		attributes:       new Set(BASE_ATTRIBUTES),
 		componentCallees: new Map(),
 	};
 }
@@ -46,8 +50,8 @@ function cloneIndex(source) {
 	for (const key of ["atoms", "components", "componentAliases", "bindTypes", "types", "phases", "labels", "attributes"]) {
 		for (const value of source[key]) result[key].add(value);
 	}
-	for (const [name, domain] of source.macros) result.macros.set(name, domain);
-	for (const [name, domain] of source.registers) result.registers.set(name, domain);
+	for (const [name, domain]  of source.macros)           result.macros.set(name, domain);
+	for (const [name, domain]  of source.registers)        result.registers.set(name, domain);
 	for (const [name, callees] of source.componentCallees) result.componentCallees.set(name, callees.slice());
 	return result;
 }
@@ -63,7 +67,7 @@ function mergeIndexes(...sources) {
 			const existing = result.macros.get(name);
 			if (!existing || domainRank(domain) >= domainRank(existing)) result.macros.set(name, domain);
 		}
-		for (const [name, domain] of source.registers) result.registers.set(name, domain);
+		for (const [name, domain]  of source.registers) result.registers.set(name, domain);
 		for (const [name, callees] of source.componentCallees) {
 			const existing = result.componentCallees.get(name) || [];
 			result.componentCallees.set(name, existing.concat(callees));
@@ -75,21 +79,21 @@ function mergeIndexes(...sources) {
 function domainFromPath(filePath) {
 	const base = path.basename(filePath.replaceAll("\\", "/")).toLowerCase();
 	if (base === "mips.h") return "cpu";
-	if (base === "gte.h") return "gte";
-	if (base === "gp.h") return "gpu";
+	if (base === "gte.h")  return "gte";
+	if (base === "gp.h")   return "gpu";
 	return null;
 }
 
 function prefixDomain(name) {
 	if (/^(?:branch_|jump_|call_)/.test(name)) return "control";
 	if (/^gte_(?!cr_)/.test(name) || name.startsWith("mac_gte_") || name.startsWith("ac_gte_")) return "gte";
-	if (/^gp[01]_/.test(name) || name.startsWith("mac_gp_") || name.startsWith("ac_gp_")) return "gpu";
+	if (/^gp[01]_/.test(name)     || name.startsWith("mac_gp_")  || name.startsWith("ac_gp_"))  return "gpu";
 	return null;
 }
 
 function collectBraceIdentifiers(tokens, openBraceIndex) {
 	const names = [];
-	let depth = 0;
+	let   depth = 0;
 	for (let tokenIndex = openBraceIndex; tokenIndex < tokens.length; tokenIndex += 1) {
 		if (tokens[tokenIndex].text === "{") depth += 1;
 		if (tokens[tokenIndex].text === "}") {
@@ -103,17 +107,17 @@ function collectBraceIdentifiers(tokens, openBraceIndex) {
 
 function resolveComponentDomains(index) {
 	const hardwareRank = { cpu: 1, gpu: 2, gte: 3, control: 4 };
-	let changed = true;
+	let   changed      = true;
 	while (changed) {
 		changed = false;
 		for (const [alias, callees] of index.componentCallees) {
-			let best = index.macros.get(alias) || "component";
+			let best     = index.macros.get(alias) || "component";
 			let bestRank = hardwareRank[best] || 0;
 			for (const callee of callees) {
 				const domain = prefixDomain(callee) || index.macros.get(callee);
-				const rank = hardwareRank[domain] || 0;
+				const rank   = hardwareRank[domain] || 0;
 				if (rank > bestRank) {
-					best = domain;
+					best     = domain;
 					bestRank = rank;
 				}
 			}
@@ -134,7 +138,7 @@ function domainRank(domain) {
 }
 
 function registerKind(name) {
-	if (/^R_[A-Za-z0-9_]+$/.test(name)) return "gpr";
+	if (/^R_[A-Za-z0-9_]+$/.test(name))              return "gpr";
 	if (/^(?:C2_|gte_cr_)[A-Za-z0-9_]+$/.test(name)) return "cop2";
 	return null;
 }
@@ -161,14 +165,15 @@ function findFunctionNameBefore(tokens, calleeTokenIndex) {
 	return null;
 }
 
-function scanSource(source, filePath) {
-	const lexical = lex(source);
-	const balanced = buildCallContexts(lexical.tokens);
-	const tokens = lexical.tokens;
-	const contexts = balanced.contexts;
-	const index = createIndex();
+function scanSource(source, filePath)
+{
+	const lexical      = lex(source);
+	const balanced     = buildCallContexts(lexical.tokens);
+	const tokens       = lexical.tokens;
+	const contexts     = balanced.contexts;
+	const index        = createIndex();
 	const declarations = new Map();
-	const domain = domainFromPath(filePath);
+	const domain       = domainFromPath(filePath);
 
 	function mark(token, role, modifiers = ["declaration"]) {
 		declarations.set(token.start, { role, modifiers });
@@ -191,7 +196,8 @@ function scanSource(source, filePath) {
 		if (!index.macros.has(alias)) index.macros.set(alias, "component");
 	}
 
-	for (let tokenIndex = 0; tokenIndex < tokens.length; tokenIndex += 1) {
+	for (let tokenIndex = 0; tokenIndex < tokens.length; tokenIndex += 1)
+	{
 		const token = tokens[tokenIndex];
 		if (token.kind !== "identifier") continue;
 
@@ -240,12 +246,14 @@ function scanSource(source, filePath) {
 			mark(token, "gprRegister", ["declaration", "tapeAuto"]);
 		}
 
-		if (token.text === "define" && tokens[tokenIndex - 1] && tokens[tokenIndex - 1].text === "#") {
+		if (token.text === "define" && tokens[tokenIndex - 1] && tokens[tokenIndex - 1].text === "#")
+		{
 			const name = tokens[tokenIndex + 1];
 			if (name && name.kind === "identifier" && name.line === token.line) {
 				if (/^(?:RegUse_|Struct_|Enum_|Union_|TypeR_|TypeV_|Relative_|Binds_)/.test(name.text)) {
 					index.types.add(name.text);
-				} else if (/^(?:ac_|mac_)/.test(name.text)) {
+				}
+				else if (/^(?:ac_|mac_)/.test(name.text)) {
 					const alias = name.text.startsWith("ac_") ? componentAlias(name.text) : name.text;
 					const rest = [];
 					for (let restIndex = tokenIndex + 2; restIndex < tokens.length && tokens[restIndex].line === name.line; restIndex += 1) {
@@ -256,7 +264,8 @@ function scanSource(source, filePath) {
 						index.macros.set(alias, prefixDomain(alias) || "component");
 						if (rest.length) index.componentCallees.set(alias, rest);
 					}
-				} else {
+				}
+				else {
 					index.macros.set(name.text, domain || "utility");
 				}
 			}

code/duffle/dsl.atom.h

@@ -44,7 +44,7 @@
  *
  *  Annotation rules
  *  ----------------
- *      1. atom_info(...) is OPTIONAL. Atoms without atom_info are silently skipped by the metaprogram.
+ *      1. atom_info(...) is optional. Atoms without atom_info are silently skipped by the metaprogram.
  *      2. If present, atom_info takes up to three sub-calls, all order-independent within the arg list:
  *           - atom_bind(Binds_X)
  *           - atom_reads(...)
@@ -160,6 +160,8 @@
  * ----------------------------------------------------------------------------*/
 #define atom_bind(binds_struct)  /* atom_bind(binds_struct) */
 
+#define Binds_(type) (tmpl(Binds,type)) // TODO(Ed): Do we want to use this?
+
 /* ============================================================================
  * atom_label / atom_offset — branch target machinery
  *
@@ -177,9 +179,3 @@
 #define atom_offset(F, T) atom_offset_ ## F ## _ ## T
 // atom_label is a pure annotation for the metaprogram's offset calculations.
 #define atom_label(name) /* atom_label anchor: name */
-
-// WIP: Atoms Assocated closely with each other to form a tape procedure. (Maybe also a phase in a procedure/pipeline?)
-
-#define AtomBundle_(name)              Struct_(tmpl(AtomBundle,name))
-#define AtomBundle_Len(name)           S_(tmpl(AtomBundle,name))/S_(MipsAtom*)
-#define AtomBundleEntry_(bundle,entry) tmpl(bundle,entry)

code/duffle/dsl.h

@@ -14,8 +14,8 @@
 #define glue(A, B)         glue_impl(A, B)
 #define tmpl(prefix, type) prefix ## _ ## type
 
-#define stringify_impl(S)        #S
-#define stringify(S)             stringify_impl(S)
+#define stringify_impl(S) #S
+#define stringify(S)      stringify_impl(S)
 
 #define VA_Sel_1( _1, ... ) _1 // <-- Of all th args passed pick _1.
 #define VA_Sel_2( _1, _2, ... ) _2 // <-- Of all the args passed pick _2.
@@ -29,8 +29,7 @@
 
 #define asm           __asm__
 
-#define A_(data)      (& data)
-#define align_(value) __attribute__((aligned (value)))             // for easy alignment
+#define A_(data)      (& (data))
 #define align_(value) __attribute__((aligned (value)))             // for easy alignment
 #define C_(type,data) ((type)(data))                               // for enforced precedence
 #define expect_(x, y) __builtin_expect(x, y)                       // so compiler knows the common path
@@ -45,7 +44,8 @@
 #define R_    restrict 
 #define V_    volatile 
 
-#pragma region Fictional //, used for intiution
+#pragma region Fictional 
+//, used for intiution
 
 #define EUB_  restrict // Execute Unit Bound:    Data is siloed in the ALU Register File. The Load/Store Unit is bypassed. (Route to Execution Unit.  Keep in registers)
 #define ISO_  restrict // Isolated Provenance:   Alternative to Exu_. Guarantees electrical memory isolation, 
@@ -83,8 +83,8 @@
 
 #define r_(ptr)        C_(T_(ptr[0])*R_, ptr) // Constrain pointer to restrict
 #define v_(ptr)        C_(T_(ptr[0])V_*, ptr) // 
-#define tr_(type, ptr) C_(type *R_, ptr)
-#define tv_(type, ptr) C_(type V_*, ptr)
+#define rt_(type, ptr) C_(type *R_, ptr)
+#define vt_(type, ptr) C_(type V_*, ptr)
 
 #define TypeR_(type)   type *R_ type ## _R  // type *restrict type_R
 #define TypeV_(type)   type V_* type ## _V  // type volatile* type_V
@@ -120,6 +120,10 @@ typedef unsigned char   TSet_(B1);
 typedef __UINT16_TYPE__ TSet_(B2); 
 typedef __UINT32_TYPE__ TSet_(B4);
 
+#define b1_(value)  C_(B1, value)
+#define b2_(value)  C_(B2, value)
+#define b4_(value)  C_(B4, value)
+
 #define u1_(value)  C_(U1, value)
 #define u2_(value)  C_(U2, value)
 #define u4_(value)  C_(U4, value)
@@ -127,6 +131,13 @@ typedef __UINT32_TYPE__ TSet_(B4);
 #define s2_(value)  C_(S2, value)
 #define s4_(value)  C_(S4, value)
 
+#define b1_r(value) C_(B1*R_, value)
+#define b2_r(value) C_(B2*R_, value)
+#define b4_r(value) C_(B4*R_, value)
+#define b1_v(value) C_(B1 V_*, value)
+#define b2_v(value) C_(B2 V_*, value)
+#define b4_v(value) C_(B4 V_*, value)
+
 #define u1_r(value) C_(U1 *R_, value)
 #define u2_r(value) C_(U2 *R_, value)
 #define u4_r(value) C_(U4 *R_, value)
@@ -183,12 +194,12 @@ def_signed_ops(le, <=)
 #undef def_generic_sop
 #endif
 
-#define alignas                             _Alignas
-#define alignof                             _Alignof
-#define byte_pad(amount, ...)               B1 glue(_PAD_, __VA_ARGS__) [amount]
-#define C_ptr(type, data)                   (C_(type*, & (data)) [0])
+#define alignas               _Alignas
+#define alignof               _Alignof
+#define byte_pad(amount, ...) B1 glue(_PAD_, __VA_ARGS__) [amount]
+#define C_ptr(type, data)     (C_(type*, & (data)) [0])
 
-#define dbg_args(...)                      __VA_ARGS__
+#define dbg_args(...) __VA_ARGS__
 
 #pragma region Control Flow & Iteration
 #define each_iter(type, iter, end)             (type iter = 0; iter < end; ++ iter)

code/duffle/gcc_asm.h

@@ -79,14 +79,6 @@
  *  Why bundle the `__asm__()` wrapper?
  *  - The integer R_T4 (= 12, via R_T4_Code) already indicates the register.
  *  - The string "$12" is derived from it via reg_str, so they cannot drift apart.
- *  - Spelling `__asm__(reg_str(R_T4_Code))` at every call site is noise.
- *
- *  tmpl defined in dsl.h (token-paste glue). 
- *  rgcc define here (gcc_asm.h) because the `__asm__` keyword is GCC-specific. 
- *  Anyone porting to a different compiler's asm dialect overrides rgcc,
- *  and the integer→string derivation in rlit can be retargeted in one place.
- *
- *  For clobber lists and asm-template strings, use the bare `rlit(R_T4_Code)`.
  * ------------------------------------------------------------------------ */
 #define rgcc(n) __asm__(rlit(n))
 

code/duffle/gen/macs.h

@@ -13,7 +13,7 @@
 //   source: C:\projects\Pikuma\ps1\code\duffle\gte.h
 //   source: C:\projects\Pikuma\ps1\code\duffle\pad.h
 //   source: C:\projects\Pikuma\ps1\code\duffle\dsl.atom.h
-//   source: C:\projects\Pikuma\ps1\code\duffle\lottes_tape.h
+//   source: C:\projects\Pikuma\ps1\code\duffle\tape.h
 //   source: C:\projects\Pikuma\ps1\code\duffle\bios.h
 //   source: C:\projects\Pikuma\ps1\code\duffle\psyq.h
 //   source: C:\projects\Pikuma\ps1\code\duffle\pad.c
@@ -35,6 +35,7 @@
  *  These do NOT yield. They are expanded inline inside Tape Atoms.
  * ---------------------------------------------------------------------------*/
 // The 'Yield' sequence for Tape Atoms (mac_yield).
+// In Forth this is considered the "NEXT" mechanism.
 #define mac_yield(...) \
 	load_word(R_AtomJmp, R_TapePtr, 0) \
 	LdSlot_ \
@@ -55,6 +56,12 @@ WORD_COUNT(mac_yield_load, 1)
 ,	BdSlot_ nop
 WORD_COUNT(mac_yield_tail, 3)
 
+/* atom_dbg_skip */
+#define mac_yield_to(code_ptr) \
+	jump_reg(code_ptr) \
+,	BdSlot_ nop
+WORD_COUNT(mac_yield_to, 2)
+
 /* atom_dbg_skip */
 #define mac_load_half_v3(tx, ty, tz, base, offset) \
 	load_half(tx, base, offset + OA_(U2,[0])) \
@@ -228,21 +235,21 @@ WORD_COUNT(mac_gte_op_cross_v3s4, 13)
 WORD_COUNT(mac_gte_store_f3, 3)
 
 /* atom_dbg_skip */
-#define mac_gte_load_tri_verts(r_vert_base, r_v0, r_v1, r_v2) \
-	shift_lleft(R_AT, r_v0, v3s2_byteoff) \
-,	add_u_self(R_AT, r_vert_base) \
+#define mac_gte_load_tri_verts(vbase, v0, v1, v2) \
+	shift_lleft(R_AT, v0, v3s2_byteoff) \
+,	add_u_self(R_AT, vbase) \
 ,	load_word(R_V0, R_AT, O_(V3_S2,x)) \
 ,	load_word(R_V1, R_AT, O_(V3_S2,z)) \
 ,	LdSlot_ gte_mv_to_data_r(R_V0, C2_VXY0) \
 ,	gte_mv_to_data_r(R_V1, C2_VZ0) \
-,	shift_lleft(R_AT, r_v1, v3s2_byteoff) \
-,	add_u_self(R_AT, r_vert_base) \
+,	shift_lleft(R_AT, v1, v3s2_byteoff) \
+,	add_u_self(R_AT, vbase) \
 ,	load_word(R_V0, R_AT, O_(V3_S2,x)) \
 ,	load_word(R_V1, R_AT, O_(V3_S2,z)) \
 ,	LdSlot_ gte_mv_to_data_r(R_V0, C2_VXY1) \
 ,	gte_mv_to_data_r(R_V1, C2_VZ1) \
-,	shift_lleft(R_AT, r_v2, v3s2_byteoff) \
-,	add_u_self(R_AT, r_vert_base) \
+,	shift_lleft(R_AT, v2, v3s2_byteoff) \
+,	add_u_self(R_AT, vbase) \
 ,	load_word(R_V0, R_AT, O_(V3_S2,x)) \
 ,	load_word(R_V1, R_AT, O_(V3_S2,z)) \
 ,	LdSlot_ gte_mv_to_data_r(R_V0, C2_VXY2) \
@@ -270,10 +277,10 @@ WORD_COUNT(mac_gte_store_g4_p3, 1)
 WORD_COUNT(mac_gte_sqr_v3, 8)
 
 /* atom_dbg_skip */
-#define mac_gte_sqr_v3s4(r_sx, r_sy, r_sz, delay_slot) \
-	gte_mv_to_data_r(r_sx, C2_IR1) \
-,	gte_mv_to_data_r(r_sy, C2_IR2) \
-,	gte_mv_to_data_r(r_sz, C2_IR3) \
+#define mac_gte_sqr_v3s4(sx, sy, sz, delay_slot) \
+	gte_mv_to_data_r(sx, C2_IR1) \
+,	gte_mv_to_data_r(sy, C2_IR2) \
+,	gte_mv_to_data_r(sz, C2_IR3) \
 ,	delay_slot \
 ,	gte_cmdw_sqr
 WORD_COUNT(mac_gte_sqr_v3s4, 5)
@@ -304,12 +311,12 @@ WORD_COUNT(mac_gte_gpf_scale, 13)
 WORD_COUNT(mac_trans_mt3s3s4, 6)
 
 /* atom_dbg_skip */
-#define mac_lzcr_round_even_half_shift(r_shift, r_mag_sq, r_mag_sq_copy) \
-	and_i(r_shift, r_shift, gte_lzcr_even_mask) \
-,	or_u(r_mag_sq_copy, r_mag_sq, 0) \
-,	li_s(        r_mag_sq, 31) \
-,	sub_s(       r_mag_sq, r_mag_sq, r_shift) \
-,	shift_aright(r_mag_sq, r_mag_sq, 1)
+#define mac_lzcr_round_even_half_shift(shift, mag_sq, mag_sq_copy) \
+	and_i(shift, shift, gte_lzcr_even_mask) \
+,	or_u(mag_sq_copy, mag_sq, 0) \
+,	li_s(        mag_sq, 31) \
+,	sub_s(       mag_sq, mag_sq, shift) \
+,	shift_aright(mag_sq, mag_sq, 1)
 WORD_COUNT(mac_lzcr_round_even_half_shift, 5)
 
 #define mac_gte_general_purpose_interopolation(to_ir0, to_ir1, to_ir2, to_ir3, fr_mac1, fr_mac2, fr_mac3, nop_slot1, nop_slot2) \

code/duffle/gen/offsets.h

@@ -10,7 +10,7 @@
 //   source: C:\projects\Pikuma\ps1\code\duffle\gte.h
 //   source: C:\projects\Pikuma\ps1\code\duffle\pad.h
 //   source: C:\projects\Pikuma\ps1\code\duffle\dsl.atom.h
-//   source: C:\projects\Pikuma\ps1\code\duffle\lottes_tape.h
+//   source: C:\projects\Pikuma\ps1\code\duffle\tape.h
 //   source: C:\projects\Pikuma\ps1\code\duffle\bios.h
 //   source: C:\projects\Pikuma\ps1\code\duffle\psyq.h
 //   source: C:\projects\Pikuma\ps1\code\duffle\pad.c
@@ -33,7 +33,7 @@ enum {
     atom_offset_example_atom_proc_skip = _atom_offset_example_atom_proc_skip,
 };
 
-// --- atom: build_normalize_v3s4 (63 words) ---
+// --- atom: normalize_v3s4 (62 words) ---
 
 #define _atom_offset_aligned_done_srav_path           3
 #define _atom_offset_srav_path_aligned_done           4

code/duffle/gp.atom.c

@@ -1,14 +1,14 @@
 #ifdef INTELLISENSE_DIRECTIVES
 #	include "dsl.h"
 #	include "gp.h"
-#	include "lottes_tape.h"
+#	include "tape.h"
 #endif
 
 ATOM_FILE_DEBUGGER_LINE_MARKER(gp_atom_c);
 
 #pragma region MACs (Mips Atom Components)
 
-FI_ Slice_MipsCode ac_gcmd_push(AtomBuilder_R ab, U4 cmd, U4 reg_transfer, U4 reg_base, U2 port)
+FI_ Slice_MipsCode ac_gcmd_push(AtomBuilder_R ab, U2 cmd, Reg reg_transfer, Reg reg_base, U2 port)
 atom_dbg_skip MipsAtomComp_Proc_(ab, {
 	mac_load_word_imm(reg_transfer, cmd),
 	store_word(       reg_transfer, reg_base, port),

code/duffle/gp.h

@@ -6,16 +6,6 @@
  *  Primitive commands: gp0_cmd_poly_f3 = 0x20    (byte opcode)
  *  Packed 32-bit cmd:  gp0_word_poly_f3(r, g, b) (32-bit, shifted)
  *
- *  Type ordering: domain?_(direction)?_action_target_modifier_type?
- *  Examples: add_ui             (add + unsigned + immediate)
- *            add_s              (add + signed, R-type implicit)
- *            shift_lleft        (shift + logical + left)
- *            shift_aright       (shift + arithmetic + right)
- *            call_reg(rs)       (call + register, $ra implicit)
- *            gte_mv_to_data_r   (gte + mv + to + data + register)
- *            gte_lw_v0_xy(base) (gte + lw + v0 + xy)
- *            load_upper_i       (load-upper + immediate, unique verb)
- *
  *  --- GPU-domain layer cake ---
  *  Every gp.h macro follows the same 4-layer composition as mips.h and gte.h:
  *    4. Semantic encoders      gp0_word_poly_f3(r,g,b)
@@ -23,9 +13,6 @@
  *    2. Per-field encoders     enc_gp0_color_r(r), enc_gp0_color_g(g), ...
  *    1. Bitfield layout consts gp0_color_red_pos = 0, gp0_color_red_width = 8
  *    0. Opcode IDs             gp0_cmd_poly_f3 = 0x20
- *
- *  Vendor mnemonics (gte_mtc2, gte_mfc2, etc.) are NOT in this header.
- *  They live in the opt-in `gp_vendor_sym.h` for users who prefer the PSYQ-style names.
  * ============================================================================ */
 
 #ifdef INTELLISENSE_DIRECTIVES
@@ -391,17 +378,15 @@ enum {
  *  Primitive structs (8 polygon variants + tag)
  *  ============================================================================
  *  Each struct follows the GPU-documented memory layout for the corresponding primitive command. 
- *  The PolyTag is the OT-link header; the rest of the struct is the primitive's body.
+ *  PolyTag is an OT-link header. Rest of the struct is the primitive's body.
  *
- *  The current working layouts match the existing demo 
- *  (floor_tri uses Poly_F3; cube_tri uses Poly_G4). 
  *  They are NOT necessarily byte-identical to the PSX-SPX reference layout. 
  *  The demo layout uses color+vertex interleaving that doesn't match the standard PSX SDK file format. 
  *  For PSX-SDK file compatibility, the textured variants (FT*, GT*) would need layout adjustments.
  * ============================================================================ */
 
 /* ---------- RGB8 (3-byte packed color) ---------- */
-typedef Struct_(RGB8) { B1 r; B1 g; B1 b; };
+typedef Struct_(RGB8) { U1 r; U1 g; U1 b; };
 #define rgb8(r,g,b) ((RGB8){r,g,b})
 
 /* ---------- PolyTag (the OT-link header; 1 word) ---------- */
@@ -431,7 +416,7 @@ typedef Struct_(PolyTag) {
 typedef Struct_(Poly_F3) {
 	U4   tag;
 	RGB8 color;
-	B1   code;
+	U1   code;
 	union {
 		struct { V2_S2 p0; V2_S2 p1; V2_S2 p2; };
 		A3_V2_S2 points;
@@ -442,7 +427,7 @@ typedef Struct_(Poly_F3) {
 typedef Struct_(Poly_F4) {
 	U4   tag;
 	RGB8 color;
-	B1   code;
+	U1   code;
 	union {
 		struct { V2_S2 p0; V2_S2 p1; V2_S2 p2; V2_S2 p3; };
 		A4_V2_S2 points;
@@ -451,18 +436,18 @@ typedef Struct_(Poly_F4) {
 
 /* ---------- Poly_G3 (Gouraud Triangle; 7 words) ---------- */
 typedef Struct_(Poly_G3) {
-	U4    tag; RGB8 c0; B1 code;
-	V2_S2 p0;  RGB8 c1; B1 pad1;
-	V2_S2 p1;  RGB8 c2; B1 pad2;
+	U4    tag; RGB8 c0; U1 code;
+	V2_S2 p0;  RGB8 c1; U1 pad1;
+	V2_S2 p1;  RGB8 c2; U1 pad2;
 	V2_S2 p2;
 };
 
 /* ---------- Poly_G4 (Gouraud Quad; 9 words) ---------- */
 typedef Struct_(Poly_G4) {
-	U4    tag; RGB8 c0; B1 code;
-	V2_S2 p0;  RGB8 c1; B1 pad1;
-	V2_S2 p1;  RGB8 c2; B1 pad2;
-	V2_S2 p2;  RGB8 c3; B1 pad3;
+	U4    tag; RGB8 c0; U1 code;
+	V2_S2 p0;  RGB8 c1; U1 pad1;
+	V2_S2 p1;  RGB8 c2; U1 pad2;
+	V2_S2 p2;  RGB8 c3; U1 pad3;
 	V2_S2 p3;
 };
 
@@ -471,7 +456,7 @@ typedef Struct_(Poly_G4) {
 typedef Struct_(Poly_FT3) {
 	U4   tag;
 	RGB8 color;
-	B1   code;
+	U1   code;
 	U4   tpage;
 	U4   clut;
 	V2_S2 p0; U1 u0; U1 v0;
@@ -483,7 +468,7 @@ typedef Struct_(Poly_FT3) {
 typedef Struct_(Poly_FT4) {
 	U4   tag;
 	RGB8 color;
-	B1   code;
+	U1   code;
 	U4   tpage;
 	U4   clut;
 	V2_S2 p0; U1 u0; U1 v0;
@@ -494,9 +479,9 @@ typedef Struct_(Poly_FT4) {
 
 /* ---------- Poly_GT3 (Gouraud Textured Triangle) ---------- */
 typedef Struct_(Poly_GT3) {
-	U4    tag; RGB8 c0; B1 code;
-	V2_S2 p0;  RGB8 c1; B1 pad1;
-	V2_S2 p1;  RGB8 c2; B1 pad2;
+	U4    tag; RGB8 c0; U1 code;
+	V2_S2 p0;  RGB8 c1; U1 pad1;
+	V2_S2 p1;  RGB8 c2; U1 pad2;
 	V2_S2 p2;
 	U4    tpage;
 	U4    clut;
@@ -507,10 +492,10 @@ typedef Struct_(Poly_GT3) {
 
 /* ---------- Poly_GT4 (Gouraud Textured Quad) ---------- */
 typedef Struct_(Poly_GT4) {
-	U4    tag; RGB8 c0; B1 code;
-	V2_S2 p0;  RGB8 c1; B1 pad1;
-	V2_S2 p1;  RGB8 c2; B1 pad2;
-	V2_S2 p2;  RGB8 c3; B1 pad3;
+	U4    tag; RGB8 c0; U1 code;
+	V2_S2 p0;  RGB8 c1; U1 pad1;
+	V2_S2 p1;  RGB8 c2; U1 pad2;
+	V2_S2 p2;  RGB8 c3; U1 pad3;
 	V2_S2 p3;
 	U4    tpage;
 	U4    clut;

code/duffle/gte.atom.c

@@ -3,7 +3,8 @@
 #	include "gen/offsets.h"
 #	include "gte.h"
 #	include "gp.h"
-#	include "lottes_tape.h"	
+#	include "tape.h"	
+#	include "math.atom.h"
 #endif
 
 ATOM_FILE_DEBUGGER_LINE_MARKER(gte_atom_c);
@@ -63,10 +64,10 @@ FI_ Slice_MipsCode ac_gte_store_f3(AtomBuilder_R ab, U4 r_primitive_cursor) atom
 })
 
 /* Words: 18; Translates indices to vertex addresses and pushes them to GTE  */
-I_ Slice_MipsCode ac_gte_load_tri_verts(AtomBuilder_R ab, U4 r_vert_base, U4 r_v0, U4 r_v1, U4 r_v2) atom_dbg_skip MipsAtomComp_Proc_(ab, {
-	shift_lleft(R_AT, r_v0, v3s2_byteoff), add_u_self(R_AT, r_vert_base), load_word(R_V0, R_AT, O_(V3_S2,x)), load_word(R_V1, R_AT, O_(V3_S2,z)), LdSlot_ gte_mv_to_data_r(R_V0, C2_VXY0), gte_mv_to_data_r(R_V1, C2_VZ0),
-	shift_lleft(R_AT, r_v1, v3s2_byteoff), add_u_self(R_AT, r_vert_base), load_word(R_V0, R_AT, O_(V3_S2,x)), load_word(R_V1, R_AT, O_(V3_S2,z)), LdSlot_ gte_mv_to_data_r(R_V0, C2_VXY1), gte_mv_to_data_r(R_V1, C2_VZ1),
-	shift_lleft(R_AT, r_v2, v3s2_byteoff), add_u_self(R_AT, r_vert_base), load_word(R_V0, R_AT, O_(V3_S2,x)), load_word(R_V1, R_AT, O_(V3_S2,z)), LdSlot_ gte_mv_to_data_r(R_V0, C2_VXY2), gte_mv_to_data_r(R_V1, C2_VZ2),
+I_ Slice_MipsCode ac_gte_load_tri_verts(AtomBuilder_R ab, Reg vbase, Reg v0, Reg v1, Reg v2) atom_dbg_skip MipsAtomComp_Proc_(ab, {
+	shift_lleft(R_AT, v0, v3s2_byteoff), add_u_self(R_AT, vbase), load_word(R_V0, R_AT, O_(V3_S2,x)), load_word(R_V1, R_AT, O_(V3_S2,z)), LdSlot_ gte_mv_to_data_r(R_V0, C2_VXY0), gte_mv_to_data_r(R_V1, C2_VZ0),
+	shift_lleft(R_AT, v1, v3s2_byteoff), add_u_self(R_AT, vbase), load_word(R_V0, R_AT, O_(V3_S2,x)), load_word(R_V1, R_AT, O_(V3_S2,z)), LdSlot_ gte_mv_to_data_r(R_V0, C2_VXY1), gte_mv_to_data_r(R_V1, C2_VZ1),
+	shift_lleft(R_AT, v2, v3s2_byteoff), add_u_self(R_AT, vbase), load_word(R_V0, R_AT, O_(V3_S2,x)), load_word(R_V1, R_AT, O_(V3_S2,z)), LdSlot_ gte_mv_to_data_r(R_V0, C2_VXY2), gte_mv_to_data_r(R_V1, C2_VZ2),
 })
 
 /* Words: 3; Stores the 3 transformed (V2_S2 screen) vertices of the
@@ -96,11 +97,11 @@ FI_ Slice_MipsCode ac_gte_sqr_v3(AtomBuilder_R ab, U4 r_sx, U4 r_sy, U4 r_sz, U4
 })
 
 /* ─── SQR FIRE — mtc2 3 GPRs into IR1/IR2/IR3, then fire SQR. ─── */
-FI_ Slice_MipsCode ac_gte_sqr_v3s4(AtomBuilder_R ab, Reg r_sx, Reg r_sy, Reg r_sz, MipsCode delay_slot)
+FI_ Slice_MipsCode ac_gte_sqr_v3s4(AtomBuilder_R ab, Reg sx, Reg sy, Reg sz, MipsCode delay_slot)
 atom_dbg_skip MipsAtomComp_Proc_(ab, {
-	gte_mv_to_data_r(r_sx, C2_IR1),
-	gte_mv_to_data_r(r_sy, C2_IR2),
-	gte_mv_to_data_r(r_sz, C2_IR3),
+	gte_mv_to_data_r(sx, C2_IR1),
+	gte_mv_to_data_r(sy, C2_IR2),
+	gte_mv_to_data_r(sz, C2_IR3),
 	delay_slot, gte_cmdw_sqr,
 })
 
@@ -157,16 +158,13 @@ FI_ Slice_MipsCode ac_trans_mt3s3s4(AtomBuilder_R ab
  *
  * Note: C2_LZCR (cop2r31) is a fixed read-only C2 data register — the caller must read it via mfc2 from C2_LZCR;
  * there is no register choice at the hardware level. Only the GPR that holds the result is caller-determined. */
-FI_ Slice_MipsCode ac_lzcr_round_even_half_shift(AtomBuilder_R ab,
-	U4 r_shift,
-	U4 r_mag_sq,
-	U4 r_mag_sq_copy)
+FI_ Slice_MipsCode ac_lzcr_round_even_half_shift(AtomBuilder_R ab, Reg shift, Reg mag_sq, Reg mag_sq_copy)
 atom_dbg_skip MipsAtomComp_Proc_(ab, {
-	and_i(r_shift, r_shift, gte_lzcr_even_mask),
-	or_u(r_mag_sq_copy, r_mag_sq, 0),
-	li_s(        r_mag_sq, 31),
-	sub_s(       r_mag_sq, r_mag_sq, r_shift),
-	shift_aright(r_mag_sq, r_mag_sq, 1),
+	and_i(shift, shift, gte_lzcr_even_mask),
+	or_u(mag_sq_copy, mag_sq, 0),
+	li_s(        mag_sq, 31),
+	sub_s(       mag_sq, mag_sq, shift),
+	shift_aright(mag_sq, mag_sq, 1),
 })
 
 FI_ Slice_MipsCode ac_gte_general_purpose_interopolation(AtomBuilder_R ab
@@ -186,60 +184,95 @@ MipsAtomComp_Proc_(ab, {
 	gte_mv_from_data_r(fr_mac3, C2_MAC3),
 })
 
-FI_ Slice_MipsCode ac_gte_mv_from_data_r_mac123(AtomBuilder_R ab
-	, Reg fr_mac1, Reg fr_mac2, Reg fr_mac3)
+FI_ Slice_MipsCode ac_gte_mv_from_data_r_mac123(AtomBuilder_R ab, Reg fr_mac1, Reg fr_mac2, Reg fr_mac3) 
 MipsAtomComp_Proc_(ab, {
 	gte_mv_from_data_r(fr_mac1, C2_MAC1),
 	gte_mv_from_data_r(fr_mac2, C2_MAC2),
 	gte_mv_from_data_r(fr_mac3, C2_MAC3),
 })
 
-
 FI_ Slice_MipsCode ac_gte_mv_from_mac123_v3s4(AtomBuilder_R ab, Reg_(V3_S4) v) MipsAtomComp_ProcMap_(ab, mac_gte_mv_from_data_r_mac123(v.x, v.y, v.z))
 
 #pragma endregion MACs (Mips Atom Components)
 
 #pragma region Atom Procs
 
-/* ─── Local copy of PSYQ's sqrtbl (1/sqrt lookup table for VectorNormal). ───
- * Source: PSYQ 4.7 libgte sqrtbl at 0x800185B4 in hello_camera.elf.
- *   objdump -s --start-address=0x800185B4 --stop-address=0x800185F4 hello_camera.elf → 192 entries × 16-bit signed, in 1.12 fixed-point (max value 0x1000 = 1.0).
- *
- * Data is identical to the libgte original (byte-for-byte verified).
+/* Normalize V3_S4 using the PSYQ/libgte reciprocal-sqrt method:
+ *   |v|² = x² + y² + z²
+ *   LZCR determines the exponent of |v|².
+ *   Round that exponent even and shift |v|² into [1, 4).
+ *   sqrtbl approximates 1/sqrt(mantissa).
+ *   GPF multiplies v by that reciprocal-sqrt mantissa.
+ *   srav_shift restores the exponent scale.
+ * Effectively: v_normalized = v * (1 / sqrt(|v|²)).
  * 
- * ─── Per-entry semantics (decoded from libgte msc02 VectorNormal) ───
- * Each entry is `1/sqrt(x)` in 1.12 fixed point (value / 4096).
- * The 192 entries span 4 octaves of the input magnitude, with 48 entries per octave:
- *   Octave 0 (entries  0- 47): mantissa in [0x8000,  0x10000)  output ~[1.000, 0.707]
- *   Octave 1 (entries 48- 95): mantissa in [0x10000, 0x20000)  output ~[0.707, 0.500]
- *   Octave 2 (entries 96-143): mantissa in [0x20000, 0x40000)  output ~[0.500, 0.354]
- *   Octave 3 (entries144-191): mantissa in [0x40000, 0x80000)  output ~[0.354, 0.251]
- * Within each octave, 8 sub-entries interpolate over the 8 fractional bits of the mantissa
- * (the byte `(0x80 | (i mod 8))` for the lower-byte of the aligned value).
- * Sampling the first value of each octave:
- *   [0]   0x1000 = 1.0000                 ; 1 / sqrt(1.0000)
- *   [48]  0x0e4f = 0.8940                 ; 1 / sqrt(1.2500)
- *   [96]  0x0d10 = 0.8164                 ; 1 / sqrt(1.5000)
- *   [144] 0x0c0a = 0.7520                 ; 1 / sqrt(1.7500)
- * And representative sub-entries within octave 0 (mantissa in [0x8000, 0x8100)):
- *   [0]   0x1000 = 1.0000                 ; 1 / sqrt(0x8000)
- *   [1]   0x0fe0 = 0.9922                 ; 1 / sqrt(0x8100)
- *   [2]   0x0fc1 = 0.9846                 ; 1 / sqrt(0x8200)
- *   [3]   0x0fa3 = 0.9773                 ; 1 / sqrt(0x8300)
- *   [4]   0x0f85 = 0.9700                 ; 1 / sqrt(0x8400)
- *   [5]   0x0f68 = 0.9629                 ; 1 / sqrt(0x8500)
- *   [6]   0x0f4c = 0.9561                 ; 1 / sqrt(0x8600)
- *   [7]   0x0f30 = 0.9492                 ; 1 / sqrt(0x8700)
+ * ─── Local port of PSYQ's sqrtbl (1/sqrt lookup table for VectorNormal). ───
+ * Source: PSYQ 4.7 libgte sqrtbl at 0x800185B4 in hello_camera.elf.
+ *   objdump -s --start-address=0x800185B4 --stop-address=0x800185F4 hello_camera.elf -> 192 entries x 16-bit signed, stored in 1.12 fixed point.
+ * Data is identical to the libgte original (byte-for-byte verified).
  *
- * The algorithm's `addi -64 / sll 1 / lh` selects the entry at `(aligned - 64) * 2` for the case where `aligned` has its top bit at bit 24.
- * After the sllv/srav pair, `aligned` always lands in `[0x80, 0x100)` 
- * (with top bit at bit 24 → after `sub $aligned - 64`, the index sits in `[0x40, 0x80) * 2 = [0x80, 0x100)` bytes = entries [64, 128) within the sqrtbl).
- * The earlier 64 entries (octave 0) are reached when the magnitude after shifting puts the top bit below bit 24 (the `sllv` branch),
- * and the load upper_halves of the table bracket the input range.
- * The later 64 entries (octaves 2-3) are the `srav` branch when the magnitude's top bit is well above bit 24.
+ * ─── Table semantics ───
+ * For table index i in [0, 192):
+ *     x      = 1 + i / 64
+ *     tbl[i] = floor(4096 / sqrt(x))
+ * Thus the table uniformly samples 1/sqrt(x) over:
+ *     x in [1.0, 4.0)
+ * at steps of 1/64, with the result represented in 1.12 fixed point (0x1000 = 1.0).
  *
- * Reproduced verbatim from libgte (verified against libpsn00b/psxgte/vector.s:100-123 — 24 rows × 8 halfwords, last entry 0x0804).
- * */
+ * Representative entries:
+ *     [  0] 0x1000 = 1.000000   ; 1 / sqrt(1.000000)
+ *     [ 16] 0x0e4f = 0.894287   ; 1 / sqrt(1.250000)
+ *     [ 32] 0x0d10 = 0.816406   ; 1 / sqrt(1.500000)
+ *     [ 48] 0x0c18 = 0.755859   ; 1 / sqrt(1.750000)
+ *     [ 64] 0x0b50 = 0.707031   ; 1 / sqrt(2.000000)
+ *     [128] 0x093c = 0.577148   ; 1 / sqrt(3.000000)
+ *     [191] 0x0804 = 0.500977   ; 1 / sqrt(3.984375)
+ *
+ * ─── How VectorNormal indexes it ───
+ * Let:
+ *     mag_sq = x*x + y*y + z*z
+ *     lzcr   = leading-zero count of mag_sq
+ * For a non-zero magnitude, libgte first rounds LZCR down to an even number:
+ *     lzcr_even = lzcr & ~1
+ *
+ * It then shifts mag_sq so that its significant bits land in one of two
+ * adjacent normalized ranges:
+ *     if lzcr_even >= 24:
+ *         aligned = mag_sq << (lzcr_even - 24)
+ *     else:
+ *         aligned = mag_sq >> (24 - lzcr_even)
+ *
+ * Because lzcr_even differs from the true LZCR by at most one bit:
+ *     raw LZCR even -> aligned in [0x80, 0x100)
+ *     raw LZCR odd  -> aligned in [0x40, 0x080)
+ * therefore:
+ *     aligned in [0x40, 0x100)
+ *
+ * Dividing this normalized integer by 64 gives exactly the table domain:
+ *     x = aligned / 64
+ *     x in [1.0, 4.0)
+ *
+ * The lookup is therefore:
+ *     index       = aligned - 0x40
+ *     byte_offset = index * sizeof(S2)
+ *     inv_len     = sqrtbl[index]
+ * or equivalently, matching the libgte instructions:
+ *     addi aligned, -64
+ *     sll  aligned, 1
+ *     lh   inv_len, sqrtbl + aligned
+ *
+ * ─── Why the domain spans [1, 4) instead of [1, 2) ───
+ * Square-root scaling depends on the parity of the exponent.
+ * Rounding LZCR to even absorbs exponent changes in pairs of bits, leaving the lookup mantissa normalized over a factor-of-four interval [1, 4).
+ *
+ * The corresponding exponent correction is retained separately as:
+ *     srav_shift = (31 - lzcr_even) >> 1
+ *
+ * After GPF multiplies the original vector components by the table's reciprocal-square-root coefficient,
+ * this shift restores the exponent scale and yields the normalized vector.
+ *
+ * Reproduced verbatim from libgte; also matches PSn00bSDK VectorNormalS _norm_table (24 rows x 8 halfwords, final entry 0x0804).
+ **/
 internal S2 const gte_normalize_sqr_tbl[192] align_(2) = {
 	0x1000, 0x0fe0, 0x0fc1, 0x0fa3, 0x0f85, 0x0f68, 0x0f4c, 0x0f30,
 	0x0f15, 0x0efb, 0x0ee1, 0x0ec7, 0x0eae, 0x0e96, 0x0e7e, 0x0e66,
@@ -267,28 +300,25 @@ internal S2 const gte_normalize_sqr_tbl[192] align_(2) = {
 	0x0820, 0x081c, 0x0818, 0x0814, 0x0810, 0x080c, 0x0808, 0x0804,
 };
 
-typedef Struct_(Binds_NormalizeV3S4) {
-	U2 src_offset;    /* offset of src V3_S4 within the BIOS scratchpad */
-	U2 dst_offset;    /* offset of dst V3_S4 within the BIOS scratchpad */
-};
-typedef Struct_(RegUse_build_normalize_v3s4) {
+typedef Struct_(Binds_normalize_v3s4) { U2 src_offset; U2 dst_offset; };
+typedef Struct_(RegUse_normalize_v3s4) {
 	union { Reg_(V3_S4) res, src; };
 	union { Reg r0, src_ptr, mac2; };
 	union { Reg r1, dst_ptr; };
-	union { Reg r2, dst_offset, mac1, v_sqr_aligned; };
-	union { Reg r3, src_offset, btarget, shift_count, sqrtbl_index; };
-	union { Reg r4, mac3, v_sqr_sum, half_shift_tmp, inv_len; };
-	union { Reg r5, lzcr, half_shift; };
+	union { Reg r2, dst_offset, mac1, v_sqr_aligned, sqrtbl_byte_offset; };
+	union { Reg r3, src_offset, align_delta, shift_count, sqrtbl_lookup; };
+	union { Reg r4, mac3, v_sqr_sum, srav_shift; };
+	union { Reg r5, lzcr_raw, lzcr_even, inv_len; };
 };
 /* ─── Full normalize (all 4 stages inline) ───
  * Generic 4-stage GTE normalize (SQR → sum+LZCR → align+sqrtbl → GPF+srav). */
-internal MipsAtom* build_normalize_v3s4(AtomArena_R aa, RegUse_build_normalize_v3s4 r)
+internal MipsAtom* normalize_v3s4(AtomArena_R aa, RegUse_normalize_v3s4 r)
 MipsAtom_Proc_(aa, {
-	load_half(r.src_offset, R_TapePtr, O_(Binds_NormalizeV3S4, src_offset)),
-	load_half(r.dst_offset, R_TapePtr, O_(Binds_NormalizeV3S4, dst_offset)),
+	load_half(r.src_offset, R_TapePtr, O_(Binds_normalize_v3s4, src_offset)),
+	load_half(r.dst_offset, R_TapePtr, O_(Binds_normalize_v3s4, dst_offset)),
 	LdSlot_ add_u(r.src_ptr, R_ScratchBase, r.src_offset),
 	LdSlot_ add_u(r.dst_ptr, R_ScratchBase, r.dst_offset),
-	LdSlot_ add_ui_self(R_TapePtr, S_(Binds_NormalizeV3S4)),
+	LdSlot_ add_ui_self(R_TapePtr, S_(Binds_normalize_v3s4)),
 
 	mac_load_v3s4(r.src, r.src_ptr, 0),
 
@@ -300,33 +330,32 @@ MipsAtom_Proc_(aa, {
 	add_u_self(        r.v_sqr_sum, r.mac1),
 	add_u_self(        r.v_sqr_sum, r.mac2),
 	gte_mv_to_data_r(  r.v_sqr_sum, C2_LZCS), GteDelay_ nop2,
-	gte_mv_from_data_r(r.lzcr,      C2_LZCR), GteDelay_ nop,
+	gte_mv_from_data_r(r.lzcr_raw,  C2_LZCR), GteDelay_ nop,
 
 	/* Stage 3: even(LZCR), half-shift, align |v|² to bit 24. */
-	mac_lzcr_round_even_half_shift(r.lzcr, r.v_sqr_sum, r.v_sqr_aligned),
-	add_si(        r.btarget, r.lzcr, -24),
-	branch_lt_zero(r.btarget, atom_offset(aligned_done, srav_path)), BdSlot_ nop, /* bltz → srav_path (LZCR <  24 path) */
+	mac_lzcr_round_even_half_shift(r.lzcr_raw, r.v_sqr_sum, r.v_sqr_aligned),
+	add_si(        r.align_delta, r.lzcr_even, -24),
+	branch_lt_zero(r.align_delta, atom_offset(aligned_done, srav_path)), BdSlot_ nop, /* bltz → srav_path (LZCR <  24 path) */
 		jump_rel(atom_offset(srav_path, aligned_done)),                             /* b → aligned_done (LZCR >= 24 path) */
-		BdSlot_ shift_lleft_var(r.v_sqr_aligned, r.v_sqr_aligned, r.btarget),
+		BdSlot_ shift_lleft_var(r.v_sqr_aligned, r.v_sqr_aligned, r.align_delta),
 		atom_label(srav_path)
 			li_s( r.shift_count, 24),
-			sub_s(r.shift_count, r.shift_count, r.lzcr),
+			sub_s(r.shift_count, r.shift_count, r.lzcr_even),
 			shift_aright_var(r.v_sqr_aligned, r.v_sqr_aligned, r.shift_count),
 	atom_label(aligned_done)
-		or_u(r.half_shift, r.half_shift_tmp, 0),
-		add_si(     r.v_sqr_aligned, r.v_sqr_aligned, -64),
-		shift_lleft(r.v_sqr_aligned, r.v_sqr_aligned, 1),
-		mac_load_word_imm(r.sqrtbl_index, & gte_normalize_sqr_tbl), add_u_self(r.sqrtbl_index, r.v_sqr_aligned),
-		load_half(r.inv_len, r.sqrtbl_index, 0),
+		add_si(     r.sqrtbl_byte_offset, r.v_sqr_aligned,    -64),
+		shift_lleft(r.sqrtbl_byte_offset, r.sqrtbl_byte_offset, 1),
+		mac_load_word_imm(r.sqrtbl_lookup, & gte_normalize_sqr_tbl), add_u_self(r.sqrtbl_lookup, r.sqrtbl_byte_offset),
+		load_half(r.inv_len, r.sqrtbl_lookup, 0),
 		LdSlot_ nop,
 
 	mac_gte_general_purpose_interopolation(r.inv_len,
 		r.src.x, r.src.y, r.src.z,
 		r.res.x, r.res.y, r.res.z,
 		GteDelay_ load_word(R_AtomJmp, R_TapePtr, 0), LdSlot_   // ac_yield: word 1
-		GteDelay_ add_ui_self(         R_TapePtr, S_(MipsCode)) // ac_yield: word 2
+		GteDelay_ add_ui_self(         R_TapePtr, S_(MipsCode)) // ac_yield: word 
 	),
-	mac_shift_aright_var_v3s4_self(r.res, r.half_shift),
+	mac_shift_aright_var_v3s4_self(r.res, r.srav_shift),
 	mac_store_v3s4(r.res, r.dst_ptr, 0),
 
 	jump_reg(R_AtomJmp), BdSlot_ nop // ac_yield: word 3-4
@@ -340,21 +369,21 @@ typedef Struct_(Binds_gte_cross_v3s4) { V3_S4* src_a; V3_S4* src_b; V3_S4* out;
 typedef Struct_(RegUse_gte_cross_v3s4) {
 	Reg_(V3_S4) a;
 	Reg_(V3_S4) b;
-	union { Reg out, t0; } x;
-	union { Reg src_a, t1, rt11; } y;
-	union { Reg src_b, t2, rt22; } z;
+	Reg out;
+	Reg src_a;
+	Reg src_b;
 };
 internal MipsAtom* gte_cross_v3s4(AtomArena_R aa, RegUse_gte_cross_v3s4 r)
 atom_info(atom_bind(Binds_gte_cross_v3s4)) MipsAtom_Proc_(aa, {
-	load_word(r.y.src_a,  R_TapePtr, O_(Binds_gte_cross_v3s4,src_a)),
-	load_word(r.z.src_b,  R_TapePtr, O_(Binds_gte_cross_v3s4,src_b)),
-	load_word(r.x.out,    R_TapePtr, O_(Binds_gte_cross_v3s4,out)),
-	LdSlot_ add_ui_self(  R_TapePtr, S_(Binds_gte_cross_v3s4)),
+	load_word(r.src_a,  R_TapePtr, O_(Binds_gte_cross_v3s4,src_a)),
+	load_word(r.src_b,  R_TapePtr, O_(Binds_gte_cross_v3s4,src_b)),
+	load_word(r.out,    R_TapePtr, O_(Binds_gte_cross_v3s4,out)),
+	LdSlot_ add_ui_self(R_TapePtr, S_(Binds_gte_cross_v3s4)),
 
-	mac_load_v3s4(r.a, r.y.src_a, 0), LdSlot_
-	mac_load_v3s4(r.b, r.z.src_b, 0), LdSlot_
+	mac_load_v3s4(r.a, r.src_a, 0), LdSlot_
+	mac_load_v3s4(r.b, r.src_b, 0), LdSlot_
 	mac_gte_op_cross_v3s4(r.a, r.b), /* RT diagonal + IR + OP + MAC read + shift */
-	mac_store_v3s4(r.a, r.x.out, 0),
+	mac_store_v3s4(r.a, r.out, 0),
 
 	mac_yield()
 })

code/duffle/gte.h

@@ -1,21 +1,10 @@
 /* ============================================================================
  *  duffle DSL Suffix Conventions
  *  ============================================================================
- *
  *  Every mnemonic in this header follows the same suffix grammar:
  *
  *  Primitive commands: gp0_cmd_poly_f3 = 0x20    (byte opcode)
  *  Packed 32-bit cmd:  gp0_word_poly_f3(r, g, b) (32-bit, shifted)
- *
- *  Type ordering: domain?_(direction)?_action_target_modifier_type?
- *  Examples: add_ui             (add + unsigned + immediate)
- *            add_s              (add + signed, R-type implicit)
- *            shift_lleft        (shift + logical + left)
- *            shift_aright       (shift + arithmetic + right)
- *            call_reg(rs)       (call + register, $ra implicit)
- *            gte_mv_to_data_r   (gte + mv + to + data + register)
- *            gte_lw_v0_xy(base) (gte + lw + v0 + xy)
- *            load_upper_i       (load-upper + immediate, unique verb)
  * ============================================================================ */
 
 #ifdef INTELLISENSE_DIRECTIVES
@@ -29,21 +18,7 @@
 /* ============================================================================
  *  gte.h — Geometry Transformation Engine (COP2) for the PS1
  * ============================================================================
- *
- *  Hand-rolled DSL for emitting GTE/MIPS instruction words from C.
- *  No GCC inline-assembly string syntax in the code body.
- *
- *  STYLE NOTES
- *  -----------
- *  - Per-field encoders are named `enc_gte_<field>(value)` and each one self-masks its argument before shifting.
- *    Mirrors the `enc_op / enc_rs / enc_rt / ...` family in mips.h.
- *  - The composite `enc_gte_cmdw(sf, mx, v, cv, lm, cmd)` is a flat OR of the per-field encoders, plus the COP2/CO base.
- *  - Pre-baked shortcuts (`gte_cmd_rtpt`, `gte_cmd_rtps`, …) are defined for the common cases so call sites read like assembly source.
- *  - All register/field values are enums (not `#define`s) so they show up in debugger symbol tables and IDE autocomplete.
- *
- *  SEE ALSO
- *  --------
- *  - mips.h: The MIPS encoder layer this builds on.
+ *  DSL for emitting GTE/MIPS instruction words from C.
  */
 
 /* C2 data registers */

code/duffle/math.atom.h

@@ -2,7 +2,7 @@
 #	include "gen/macs.h"
 #	include "gen/offsets.h"
 #	include "math.h"
-#	include "lottes_tape.h"
+#	include "tape.h"
 #endif
 
 ATOM_FILE_DEBUGGER_LINE_MARKER(math_atom_c);

code/duffle/memory.h

@@ -5,7 +5,7 @@
 
 #define MEM_ALIGNMENT_DEFAULT  4
 
-#define assert_bounds(point, start, end) for(;0;){ \
+#define assert_bounds(point, start, end) do{ \
 	assert((start) <= (point)); \
 	assert((point) <= (end));   \
 } while(0)
@@ -23,10 +23,10 @@ FI_ void mem_bump(U4 cap, U4*R_ used, U4 amount) {
 	used[0] += amount; 
 }
 
-FI_ U4 mem_copy            (U4 dest, U4 src,   U4 len) { return (U4)(__builtin_memcpy ((void*)dest, (void const*)src,   len)); }
-FI_ U4 mem_copy_overlapping(U4 dest, U4 src,   U4 len) { return (U4)(__builtin_memmove((void*)dest, (void const*)src,   len)); }
-FI_ U4 mem_fill            (U4 dest, U4 value, U4 len) { return (U4)(__builtin_memset ((void*)dest, (int)        value, len)); }
-FI_ B4 mem_zero            (U4 dest,           U4 len) { if(dest == 0){return false;} mem_fill(dest, 0, len); return true; }
+FI_ U4 mem_copy            (U1_R dest, U1_R src, U4 len) { return (U4)(__builtin_memcpy ((void*)dest, (void const*)src,   len)); }
+FI_ U4 mem_copy_overlapping(U1*  dest, U1*  src, U4 len) { return (U4)(__builtin_memmove((void*)dest, (void const*)src,   len)); }
+FI_ U4 mem_fill            (U1_R dest, U4 value, U4 len) { return (U4)(__builtin_memset ((void*)dest, (int)        value, len)); }
+FI_ B4 mem_zero            (U1_R dest,           U4 len) { if(dest == 0){return false;} mem_fill(dest, 0, len); return true; }
 
 #pragma region DAG
 
@@ -58,31 +58,30 @@ typedef Struct_(Str8)        { UTF8* ptr; U4 len; };
 typedef Struct_(Slice_Str8)  { Str8* ptr; U4 len; };
 #define slit(string_literal) (Str8){ (UTF8*) string_literal, S_(string_literal) - 1 }
 
-typedef Struct_(Slice) { B1* ptr; U4 len; }; // Untyped Slice (byte-addressable; .len in elements)
-FI_ Slice slice_ut_(U4 ptr, U4 len) { return (Slice){(B1*)ptr, len}; }
+typedef Struct_(Slice) { U1* ptr; U4 len; };
+FI_ Slice slice_ut_(U1* ptr, U4 len) { return (Slice){ptr, len}; }
 
 #define Slice_(type)       Struct_(tmpl(Slice,type)) { type* ptr; U4 len; }
-typedef Slice_(B1);
 #define slice_assert(s)    do { assert((s).ptr != 0); assert((s).len > 0); } while(0)
-#define slice_end(slice)   ((slice).ptr + S_slice(slice) / S_(B1))  /* byte-ptr arithmetic; .len is in elements per slice convention */
+#define slice_end(slice)   ((slice).ptr + S_slice(slice))
 #define S_slice(s)         ((s).len * S_((s).ptr[0]))
 
-#define slice_ut(ptr,len)  slice_ut_(u4_(ptr),     u4_(len))
-#define slice_ut_arr(a)    slice_ut_(u4_(a),       S_(a))
-#define slice_to_ut(s)     slice_ut_(u4_((s).ptr), S_slice(s))
+#define slice_ut(ptr,len)  slice_ut_(C_(U1*,ptr),    u4_(len))
+#define slice_ut_arr(a)    slice_ut_(C_(U1*,a),       S_(a))
+#define slice_to_ut(s)     slice_ut_(C_(U1*,(s).ptr), S_slice(s))
 
 #define slice_iter(container, iter)     (T_((container).ptr) iter = (container).ptr; iter != slice_end(container); ++ iter)
 #define slice_arg_from_array(type, ...) & (tmpl(Slice,type)) { .ptr = Array_decl(type,__VA_ARGS__), .len = Array_len( Array_decl(type,__VA_ARGS__)) }
 #define slice_from_array(type, array)     (tmpl(Slice,type)) { .ptr = array, .len = Array_len(array) }
 
-FI_ void slice_zero_(Slice s) { slice_assert(s); mem_zero(u4_(s.ptr), s.len); }
+FI_ void slice_zero_(Slice s) { slice_assert(s); mem_zero(s.ptr, s.len); }
 #define  slice_zero(s)        slice_zero_(slice_to_ut(s))
 
 FI_ void slice_copy_(Slice dest, Slice src) {
 	assert(S_slice(dest) >= S_slice(src));
 	slice_assert(dest);
 	slice_assert(src);
-	mem_copy(u4_(dest.ptr), u4_(src.ptr), S_slice(src));
+	mem_copy(dest.ptr, src.ptr, S_slice(src));
 }
 #define slice_copy(dest, src) do {  \
 	static_assert(T_same(dest, src)); \
@@ -95,6 +94,7 @@ FI_ Slice slice_bump(U4_R used, U4 start, U4 len, U4 amount) {
 	return slice_ut(ptr, amount);
 }
 
+typedef Slice_(B1);
 typedef Slice_(U1);
 typedef Slice_(U4);
 
@@ -117,7 +117,7 @@ I_  Slice farena_push(FArena_R arena, U4 amount, Opt_farena o) {
 	U4 to_commit = align_pow2(desired, o.alignment ?  o.alignment : MEM_ALIGNMENT_DEFAULT);
 	U4 ptr       = arena->start + arena->used;
 	mem_bump(arena->capacity, & arena->used, to_commit);
-	return (Slice){ (B1*)ptr, to_commit };
+	return (Slice){ (U1*)ptr, to_commit };
 }
 FI_ void farena_reset (FArena_R arena) { arena->used = 0; }
 FI_ void farena_rewind(FArena_R arena, U4 save_point) {
@@ -134,8 +134,7 @@ FI_ U4 farena_unused_start(FArena arena) { return arena.start + arena.used; }
 
 #pragma region BIOS Scratchpad
 /* BIOS scratchpad location. 1 KB at 0x1F800000.
- * TapeHostFrame occupies the final 44 bytes while tape code executes.
- * Atom scratch is bounded by the TapeHostFrame_Loc declaration in lottes_tape.h. */
+ * TapeHostFrame occupies the final 44 bytes while tape code executes. */
 enum {
 	Scratchpad_Loc = 0x1F800000,
 	Scratchpad_Len = 0x400,                             /* 1 KB */

code/duffle/mips.atom.c

@@ -3,7 +3,7 @@
 #		include "gen/offsets.h"
 #	include "bios.h"
 #	include "mips.h"
-#	include "lottes_tape.h"
+#	include "tape.h"
 #endif
 
 ATOM_FILE_DEBUGGER_LINE_MARKER(mips_atom_c);

code/duffle/mips.h

@@ -80,14 +80,9 @@ enum {
  *  Every R_* enum below has a parallel R_*_Code `#define` so that the preprocessor can stringify the integer
  *  (e.g. for asm clobber lists and register-variable declarations via `rgcc(R_X)`).
  *  The enum value is bound to the `#define` so the two forms cannot drift apart.
- *
- *  Only registers that get stringified need a `_Code` form; the rest are plain enum values.
- *  If you need to add a new one, follow the pattern:
- *      #define R_T7_Code 15
- *      R_T7 = R_T7_Code,     // in the enum
- *
+ * 
  *  User code should always reference the enum form (`R_T4`) at arithmetic sites and let
- *  `rlit(R_T4_Code)` / `rgcc(R_T4)` handle the stringify cases — never write the bare number `12`.
+ *  `rlit(R_T4_Code)` / `rgcc(R_T4)` handle the stringify cases
  * ============================================================================ */
 #define R_0_Code    0
 #define R_AT_Code   1

code/duffle/pad.atom.c

@@ -3,7 +3,7 @@
 #		include "gen/offsets.h"
 #	include "mips.h"
 #	include "dsl.atom.h"
-#	include "lottes_tape.h"
+#	include "tape.h"
 #	include "pad.h"
 #endif
 

code/duffle/pad.c

@@ -6,16 +6,10 @@
 #	include "pad.h"
 #endif
 
-/* Uses ONE 8-byte frame allocated via the compiler's standard prologue.
+/* Uses an 8-byte frame allocated via the compiler's standard prologue.
  * 4 wasted-arg words for B(12h) InitPAD2 are at [SP+0..15] but are not explicitly allocated.
  * Compiler handles the MIPS O32 "wasted stack" convention for us by treating the B-call as a 4-arg call.
- *
- * The buffer pointers are passed as arguments so the compiler keeps them in callee-saved registers;
- * The B(12h) asm volatile block does NOT clobber those registers (it clobbers only the volatile GPRs + B-table arg registers explicitly).
- * The C-level writes after the call re-load the pointers from their callee-saved homes.
- *
- * The clobber list for both B-calls names the full BIOS destroy set documented in kernelbios.md:167-174 (R1..R15, R24..R25, R31, HI/LO).
- * The kernel-ABI "volatile GPRs" subset is clb_mem_drain; the rest of the destroy set is enumerated explicitly here. */
+ */
 NI_ void pad_bios_init_start(PadBiosRaw* raw0, PadBiosRaw* raw1)
 {
 	/* Pin raw0 + raw1 to $a0 + $a1 via rgcc; the B(12h) call uses these directly.
@@ -24,9 +18,6 @@ NI_ void pad_bios_init_start(PadBiosRaw* raw0, PadBiosRaw* raw1)
 	register PadBiosRaw* p1 rgcc(R_A1) = raw1;
 	(void)p0; (void)p1;
 
-	// TODO(Ed): Properly annotate the raw values in the inline asm instructions.
-	// Use enums.
-
 	/* B(12h) InitPAD2(raw0, 0x22, raw1, 0x22)
 	 *   $a0 = raw0 (rgcc-bound; survives the sequence below)
 	 *   $a1 = raw1 (preserved into $a2 before $a1 is overwritten)

code/duffle/pad.h

@@ -4,10 +4,7 @@
 #	include "math.h"
 #endif
 
-/* PSX button bit positions — 1:1 with PSX-SPX docs at docs/psx-spx/docs/controllersandmemorycards.md:405-421.
- * Wire is active-low (0 = pressed).
- * The decoder atom computes buttons = (~raw_buttons) & 0xFFFF; 
- * active-low-to-active-high inversion is applied bit-by-bit. */
+// PSX button bit positions: PSX-SPX docs/psx-spx/docs/controllersandmemorycards.md:405-421.
 typedef Enum_(U2, PadBtns) {
 	Bit_(Pad_Select,    0),
 	Bit_(Pad_L3,        1),
@@ -62,11 +59,7 @@ typedef Enum_(U4, PadStatus) {
 	PadStatus_Invalid,
 };
 
-/* Distinct from the game-facing PadStatus enum: PadRawStatus_Ok and PadRawStatus_Timeout are raw BIOS values;
- * PadStatus_* are game-facing post-decode states. PadUnknownId_Sentinel is written by the decoder 
- * when the controller id does not match any known controller type.
- * PadAxisCentered_Word: Four-byte 0x80 pattern used to clear / center 
- * four byte axes at PadState.left_x through PadState.right_y. */
+// Distinct from the game-facing PadStatus enum: PadRawStatus_Ok and PadRawStatus_Timeout are raw BIOS values
 typedef Enum_(U1, PadRawStatus) {
 	PadRawStatus_Ok      = 0x00,
 	PadRawStatus_Timeout = 0xFF,

code/duffle/psyq.h

@@ -104,7 +104,7 @@ void gte_matrix_set_translation(MT3_S2S4* mat) asm("SetTransMatrix");
 
 // Einheit, Metrication to unit vector. "Normalization", not Orthogonal "Normal, Normalis". Directionalization.
 // RGA(Lengyel): Normalize the bulk of a zero-weight direction. This is not finite-point unitization (which forces w=1).
-S4 normalize_v3s4(V3_S4* v0, V3_S4* v1) asm("VectorNormal");
+S4 psy_normalize_v3s4(V3_S4* v0, V3_S4* v1) asm("VectorNormal");
 
 // RGA(Lengyel): Apply the matrix expansion of a rigid transformation.
 // Motor antiproduct is equivalent for unitized points; LA form is what GTE consumes.

code/duffle/tape.h

@@ -13,56 +13,32 @@
 
 #pragma region Tape Drive
 /* -----------------------------------------------------------------------------------------------------------
- *  TAPE DRIVE ABI
+ *  THREADED ATOMS - TAPE EXECUTION & ABI
+ *  _________
+ * |   ___   |         
+ * |  o___o  |              ,-----<-----.
+ * |__/___\__|              V           ^
+ *          \_[Enter]_[A]->[A]->[A]->[A(B)]->[A]->[Exit]
  * -----------------------------------------------------------------------------------------------------------
- * Note(Ed): One of the main purposes of this codebase is to help me learn this, 
- * as such the information below may not* be entirely realized or finalized conceptually.
- * -----------------------------------------------------------------------------------------------------------
- * This ABI and its associated legos were directly inspired by researching the work of 
- * Timothy Lottes and Onat Türkçüoğlu; along with many others. It's the simplest bootstrap of a 
- * directly executed chain of assemby arrays (Atoms) that terminate with a yield sequence to the next atom.
- * These eventually lead to a terminal atom for the tape which is defined below as "tape_exit".
+ * This ABI and its associated legos were directly inspired by researching the work of Timothy Lottes and
+ * Onat Türkçüoğlu; Forth, threaded code system, and various other people or programming techniques.
+ * 
+ * The setup is simple: 
+ * A tape is a linear stream containing addresses of directly executable native-code fragments ("Atoms").
+ * Most atoms terminate in a small yield sequence which loads the next atom address from the tape.
+ * It's a runtime composed of directly executed native machine-code sequences (Atoms) that usually terminate 
+ * in a yield sequence to the next atom. These eventually lead to a terminal atom for the tape
+ * which is defined below as "tape_exit". Traditionally referred to as Direct Threaded Execution.
  * 
  * It behaves as one of the simplest runtime harnesses ontop of a host-enviornment's execution engine 
  * to author and compose programs with. From here various conventions can be further applied. 
- * To make things easier  to understand it may be better to focus on what this ABI does not have. 
- * It does not have have any branching within the tape but relative branches within atoms or between atoms.
- * Branching nearly is always downstream. Automatic stack usage is non-existent.
- * Push/Pop, FIFO, or Arena/Bump data structures are used by atoms explicitly.
- * In it's current form with the C11 macro DSL, the user also has fullfill manual register allocation per atom.
- * 
- * One of the remarkable things about utilizing this ABI is its essentially interopable with CPUs, GPUs, FPGA, 
- * or, basically anything from the 5th generation consoles and onward.
- * The ABI directly reflects how all computational hardware must be architected in order to execute 
- * digital logic effectively on current era tech.
- * On the PS1 we don't have access to a few features like multi-threading, speculative execution, or L3 cache;
- * but, we can set the foundation for legoing whats required for eventually expanding this ABI's paradigm 
- * and core atoms to take those newer hardware features into account. For example, you can easily expand 
- * this to support wave-based execution model on a PS2 or PS3. Not having a stack or 
- * automatic register allocation means the user cannot ignore excessive argument shuffle across workload or
- * waves and thier phases. Crossing ABI boundaries to other runtimes that do has obviouss penalties.
- * 
- * Learning data-oriented code becomes a natural progression. Your not fighting a stack-based procedural 
- * paradigm that wants to argument shuffle. There is no ambiguity due to the lack of constraints, for example,
- * on how the user may "call" a procedure in traditional random dispatch runtimes. The user does have to
- * hammer down "rules" or patterns for massaging the compiler to dissolve those call frames; just to get 
- * the asesmbly into its desired form. The form is obvious, and once the user gets to author these compoonents
- * it becomes a game of tetris.
- * 
- * Another feature is this ABI is very compatible with bootstrapping and developing simple toolchains built off
- * of bit-packed annotated command streams the user can directly author, maintatain, and immediately execute.
- * That being like a color forth, or maybe something more familar like an immediate mode library 
- * for various systems such as GUIs. This can make the tetris less of a chore with some helpful policy
- * generation for allocation of registers, helping to choose resuable components, designing DSL on the fly, etc.
- * -----------------------------------------------------------------------------------------------------------
- * TODO(Ed): We need pretty ascii diagrams and proper guides, articles, etc.
- * -----------------------------------------------------------------------------------------------------------
- * For now this ideation has just started functioning. I'm abusing C11 & a lua metaprogram to help establish 
- * a hybrid toolchain to ideate on a traditional text-based authoring UX for this paradigm.
- * If pcsx-redux provides viable hot-reload and persistent data storage beyond save-states
- * (just copying ram to filesystem), I can author a color forth to mess around with.
- * With either an editor in-emulator or on the actual machine itself. Assembly is tedius,
- * but I think this codebase most likely has a pretty ergonomic flavor worst case...
+ * To make things easier to understand it may be better to focus on what this ABI does not have. 
+ * The tape itself does not have have any branching behavior.
+ * Branches, loops, skips, or other control-flow policies must be implemented explicitly by atoms.
+ * Push/Pop, FIFO, or Arena/Bump data structures are utilized by atoms explicitly.
+ * There is no implicit call-stack, return stack, or per-atom stack-frame.
+ * The user must also explictly handle register allocation per atom (by default). 
+ * However they could procedurally automate it using metaprogramming functionality.
  * */
 /* Register Allocation Info */
 enum {
@@ -99,8 +75,8 @@ enum {
 	R_Atom7  = R_T7,
 	R_Atom8  = R_T8,
 	R_Atom9  = R_T9,
-	R_Atom10 = R_V0, // Tend to be used with gte DMAs
-	R_Atom11 = R_V1, // Tend to be used with gte DMAs
+	R_Atom10 = R_V0, // Tend to be used with gte moves
+	R_Atom11 = R_V1, // Tend to be used with gte moves
 	R_Atom12 = R_A0,
 	R_Atom13 = R_A1,
 	R_Atom14 = R_A2,
@@ -121,7 +97,7 @@ typedef U2 Reg; // Register parameter used with atom or atom component procedure
 typedef U4 const MipsCode; // Underlying type to mips asm words.
 typedef Slice_(MipsCode);
 
-typedef U4 const MipsAtom; // Underlying type to a mips atom defnition 
+typedef U4 const MipsAtom; // Underlying type to a mips atom definition 
 typedef Slice_(MipsAtom);
 
 // Sometimes a user will define a bundle of atoms that represent a procedure of work as:
@@ -163,6 +139,12 @@ typedef Slice_(MipsAtom);
 // Used for trivial mappings from one atom component proc to the command of a more baser (meant for type-mapping)
 #define MipsAtomComp_ProcMap_(ab, base_command) atom_dbg_skip MipsAtomComp_Proc_(ab, {base_command })
 
+// WIP: Atoms Assocated closely with each other to form a tape procedure. (Maybe also a phase in a procedure/pipeline?)
+
+#define AtomBundle_(name)              Struct_(tmpl(AtomBundle,name))
+#define AtomBundle_Len(name)           S_(tmpl(AtomBundle,name))/S_(MipsAtom*)
+#define AtomBundleEntry_(bundle,entry) tmpl(bundle,entry)
+
 /* Line-table anchor: gcc only adds a file to the .debug_line file table when the contains line-numbered content.
 	Files containing only atoms and atom components.
    Place `ATOM_FILE_LINE_MARKER();` once at file scope in any `.atom.c` that defines atoms.
@@ -170,7 +152,7 @@ typedef Slice_(MipsAtom);
 	 The constant is in `.rodata` so the linker may eliminate it. */
 #define ATOM_FILE_DEBUGGER_LINE_MARKER(file_name) internal U4 const tmpl(atom_file_debugger_line_marker,file_name) = 0
 
-typedef Slice_MipsAtom Tape;
+typedef Struct_(Tape) { union { MipsAtom* ptr; U4* inlaid_data; }; U4 len; };
 
 typedef Struct_(TapeHostFrame) {
 	U4 s0;
@@ -236,22 +218,27 @@ FI_ void tape_run(Tape tape) { C_(TapeEntryFn*, tape_enter)(tape.ptr); }
 // Procedural authoring of tapes:
 typedef Relative_(FArena) Struct_(TapeBuilder) { U4 ptr; U4 capacity; U4 used; };
 FI_ void        tb_init(TapeBuilder* tb, FArena* arena) { tb->ptr = arena->start; tb->used = 0; }
-FI_ TapeBuilder tb_make_old(             FArena* arena) { return (TapeBuilder){ arena->start, 0 }; }
 FI_ TapeBuilder tb_make(Slice mem) { return (TapeBuilder){ u4_(mem.ptr), mem.len, 0 }; }  /* capacity in elements (matches used units) */
 
 FI_ void tb_emit(TapeBuilder* tb, MipsAtom* atom) { u4_r(tb->ptr)[tb->used] = u4_(atom); ++ tb->used; }
 FI_ void tb_data(TapeBuilder* tb, U4        data) { u4_r(tb->ptr)[tb->used] = u4_(data); ++ tb->used; }
 #define tb_emit_(atom)        tb_emit(& tb, atom)
-#define tb_data_(field, data) tb_data(& tb, u4_(data))
 
-FI_ void tb_emit_bundle(TapeBuilder_R tb, Slice_MipsAtom atoms) { mem_copy(u4_(tb->ptr), u4_(atoms.ptr), S_slice(atoms)); tb->used += atoms.len; }
+FI_ void tb_bind(TapeBuilder* tb, Slice data) { mem_copy(b1_r(tb->ptr + tb->used * S_(MipsCode)), data.ptr, data.len); tb->used += data.len / S_(MipsCode); }
+#define tb_bind_(tb,type,...) tb_bind(tb, (Slice){ (U1*)(& (type){__VA_ARGS__}), S_(type) }); static_assert(S_(type) % S_(MipsCode) == 0)
 
 FI_ Tape tb_end  (TapeBuilder* tb) { tb_emit(tb,tape_exit); return (Tape){ C_(U4*,tb->ptr), tb->used }; }
 FI_ Tape tb_slice(TapeBuilder  tb) {                        return (Tape){ C_(U4*,tb.ptr),  tb.used }; }
-#define tb_scope(tb)     for(U4 tbs_once=0;tbs_once==0;++tbs_once,tb_emit(tb,tape_exit))
+#define tb_scope(tb) for(U4 tbs_once=0;tbs_once==0;++tbs_once,tb_emit(tb,tape_exit))
 
 FI_ void tb_scope_run_end(TapeBuilder* tb) { tb_emit(tb,tape_exit); tape_run(tb_slice(tb[0])); }
 #define tb_scope_run(tb) for(U4 tbs_once=0;tbs_once==0;++tbs_once,tb_scope_run_end(tb))
+
+
+// NOTE(Ed): Wip still ideating convention. Possibly will never use a composite.
+#define tb_emit_wbind_(tb,atom,...)       tb_emit(tb,atom); tb_bind_(tb,tmpl(Binds,atom),__VA_ARGS__)
+#define tb_emit_wbind2_(tb,atom,type,...) tb_emit(tb,atom); tb_bind_(tb,type,__VA_ARGS__)
+
 #pragma endregion Tape Drive
 
 #pragma region Macro Mips Atom Components
@@ -260,7 +247,8 @@ FI_ void tb_scope_run_end(TapeBuilder* tb) { tb_emit(tb,tape_exit); tape_run(tb_
  *  These do NOT yield. They are expanded inline inside Tape Atoms.
  * ---------------------------------------------------------------------------*/
 
-// The 'Yield' sequence for Tape Atoms (mac_yield).
+// The 'Yield' sequence for Tape Atoms (mac_yield). 
+// In Forth this is considered the "NEXT" mechanism.
 
 atom_dbg_skip MipsAtomComp_(ac_yield) {
 	load_word(R_AtomJmp, R_TapePtr, 0), LdSlot_
@@ -289,7 +277,7 @@ typedef Relative_(FArena) Struct_(AtomBuilder) { U4 start; U4 capacity; U4 used;
 FI_ void atombuilder_push(AtomBuilder_R ab, Slice_MipsCode code) {
 	assert(ab->capacity - ab->used - code.len);
 	U4 dest = ab->start + ab->used * S_(MipsCode); U4 size = S_slice(code);
-	mem_copy(dest, u4_(code.ptr), size); ab->used += size;
+	mem_copy(b1_r(dest), b1_r(code.ptr), size); ab->used += size;
 }
 #define atombuilder_push_mac(ab, mac) atombuilder_push(ab, slice_arg_from_array(Slice_MipsCode, mac))
 
@@ -313,7 +301,7 @@ FI_ AtomArena atomarena_make(Slice mem) { AtomArena a; atomarena_init(& a, mem);
 FI_ MipsAtom* atomarena_push(AtomArena_R aa, Slice_MipsCode code) {
 	assert(aa->capacity - aa->used - code.len);
 	U4 dest = atomarena_unused_start(aa[0]); U4 size = S_slice(code);
-	mem_copy(dest, u4_(code.ptr), size); aa->used += size;
+	mem_copy(b1_r(dest), b1_r(code.ptr), size); aa->used += size;
 	return C_(MipsAtom*, dest);
 }
 FI_ void atomarena_reset(AtomArena_R aa) { aa->used = 0; }
@@ -344,24 +332,14 @@ internal Reg const regfile_alloc_order[] = {
 	R_T8, R_T9,
 };
 
-typedef Struct_(RegFile) {
-	A2_U2 GPR;
-	A2_U2 GTE;
-};
+typedef Struct_(RegFile) { A2_U2 GPR; };
 #define regfile(pin_mask) {.GPR={u4_lo(pin_mask), u4_hi(pin_mask)} }
 FI_ void regfile_init(RegFile_R rf) {
-	/* pack the 32-bit ABI mask into the two U2s */
-	rf->GPR[0] = u4_lo(regfile_abi_mask);
-	rf->GPR[1] = u4_hi(regfile_abi_mask);
-	rf->GTE[0] = rf->GTE[1] = 0;
+	rf->GPR[0] = u4_lo(regfile_abi_mask); rf->GPR[1] = u4_hi(regfile_abi_mask);
 }
 FI_ RegFile regfile_make(void) { RegFile rf; regfile_init(& rf); return rf; }
 
-typedef Struct_(RegFile_RInfo) {
-	U2_R section;
-	U2   mask;
-	B2   occupied;
-};
+typedef Struct_(RegFile_RInfo) { U2_R section; U2 mask; B2 occupied; };
 FI_ RegFile_RInfo regfile_rinfo(A2_U2 file, Reg r_id) {
 	U2   s_id     = r_id >> 4;
 	U2_R section  = & file[s_id];
@@ -371,19 +349,11 @@ FI_ RegFile_RInfo regfile_rinfo(A2_U2 file, Reg r_id) {
 }
 FI_ Reg regfile__alloc_helper(A2_U2 file, Reg r_id) {
 	Reg result = 0; RegFile_RInfo info = regfile_rinfo(file, r_id);
-	if (info.occupied == false) {
-		info.section[0] |= info.mask;
-		result           = r_id;
-	}
+	if (info.occupied == false) { info.section[0] |= info.mask; result = r_id; }
 	return result; 
 }
-/* regfile_alloc picks the next free GPR from regfile_alloc_order.
- * The table is the first-fit allocation order: T0..T7, V0..V1, A0..A3,
- * S0..S7, T8..T9. The 24 entries leave room for the tape program to use
- * any of them while R0, R1, R26-R31 remain reserved. */
-I_ Reg regfile_alloc(RegFile_R rf) {
-	Reg allocated = 0;
-	for index_iter(U4, r_id, R_V0, <, R_T9) {
+I_ Reg regfile_alloc(RegFile_R rf) { 
+	Reg allocated = 0; for index_iter(U4, r_id, R_V0, <, R_T9) {
 		allocated = regfile__alloc_helper(rf->GPR, r_id);
 		Jmp_nZero_(allocated,resolved);
 	}
@@ -392,13 +362,12 @@ resolved: return allocated;
 }
 FI_ Reg regfile_pin(RegFile_R rf, Reg r_id) {
 	RegFile_RInfo info = regfile_rinfo(rf->GPR, r_id);
-	assert(info.occupied == false);
+	assert(info.occupied == false); 
 	info.section[0] |= info.mask;
 	return r_id;
 }
 FI_ void regfile_pin_mask(RegFile_R rf, U4 mask) {
-	B4 occupied = u4_r(rf->GPR)[0] & mask;
-	assert(occupied == false);
+	B4 occupied = u4_r(rf->GPR)[0] & mask; assert(occupied == false);
 	u4_r(rf->GPR)[0] |= mask;
 }
 FI_ void regfile_free_mask(RegFile_R rf, U4 mask) {
@@ -406,21 +375,23 @@ FI_ void regfile_free_mask(RegFile_R rf, U4 mask) {
 	u4_r(rf->GPR)[0] &= ~mask;
 }
 FI_ void regfile_free_reg(RegFile_R rf, Reg r_id) {
-	/* never free the ABI set */
-	if (regfile_abi_mask & (1u << r_id)) return;
+	if (regfile_abi_mask & (1u << r_id)) return; // never free the ABI set
 	RegFile_RInfo info = regfile_rinfo(rf->GPR, r_id);
 	info.section[0] &= ~info.mask;
 }
-FI_ void regfile_reset(RegFile_R rf) {
-	rf->GPR[0] = u4_lo(regfile_abi_mask);
-	rf->GPR[1] = u4_hi(regfile_abi_mask);
-}
-FI_ void regfile_reset_to_mask(RegFile_R rf, U4 mask) {
-	rf->GPR[0] = u4_lo(mask);
-	rf->GPR[1] = u4_hi(mask);
-}
+FI_ void regfile_reset        (RegFile_R rf)          { rf->GPR[0] = u4_lo(regfile_abi_mask); rf->GPR[1] = u4_hi(regfile_abi_mask); }
+FI_ void regfile_reset_to_mask(RegFile_R rf, U4 mask) { rf->GPR[0] = u4_lo(mask);             rf->GPR[1] = u4_hi(mask); }
 #pragma endregion RegFileArena (Register File Allocator)
 
+#pragma region Mips Atom Components (Procedures)
+
+// For doing direct-chaining of "atoms or fragments".
+FI_ Slice_MipsCode ac_yield_to(AtomBuilder_R ab, Reg code_ptr) atom_dbg_skip MipsAtomComp_Proc_(ab, {
+	jump_reg(code_ptr), BdSlot_ nop,
+})
+
+#pragma endregion Mips Atom Components (Procedures)
+
 #pragma region Mips Atom Procs
 /* RegUse structs are a convention to organize register allocations for a mips atom procedure.
 	Unlike the usual enum-based declarations, they provide a namespaced scope and have view types via union declarations. */
@@ -441,7 +412,6 @@ FI_ void regfile_reset_to_mask(RegFile_R rf, U4 mask) {
 		add_si(r.t1.view_3, r.usual_modifiable, 10),
 		mac_yield(),
 	})
-
 #pragma endregion Mips Atom Procs
 
 #pragma region Baked Mips Atoms

code/hello_camera/hello_camera.atom.c

@@ -3,7 +3,7 @@
 #		include "duffle/gen/macs.h"
 #		include "duffle/gen/offsets.h"
 #	include "duffle/dsl.atom.h"
-#	include "duffle/lottes_tape.h"
+#	include "duffle/tape.h"
 #	include "duffle/mips.h"
 #	include "duffle/gte.h"
 #	include "duffle/gp.h"
@@ -11,8 +11,8 @@
 #	include "duffle/word_count.metadata.h"
 #	include "duffle/psyq.h"
 #	include "duffle/math.atom.h"
-#	include "duffle/mips.atom.c"
 #	include "duffle/gte.atom.c"
+#	include "duffle/mips.atom.c"
 #	include "duffle/gp.atom.c"
 #	include "duffle/psyq.atom.c"
 #		include "gen/offsets.h"
@@ -21,7 +21,7 @@
 #	include "hello_camera.h"
 #endif
 
-ATOM_FILE_DEBUGGER_LINE_MARKER(hello_joypad_atom_c);
+ATOM_FILE_DEBUGGER_LINE_MARKER(hello_camera_atom_c);
 
 #pragma region MACs (Mips Atom components)
 
@@ -136,7 +136,7 @@ atom_info(atom_bind(Binds_ResolveLookAtSub)) MipsAtom_Proc_(aa, {
 	load_word(r.target_ptr, R_TapePtr, O_(Binds_ResolveLookAtSub,target)),
 	load_word(r.eye_ptr,    R_TapePtr, O_(Binds_ResolveLookAtSub,eye)),
 	load_word(r.up_in_ptr,  R_TapePtr, O_(Binds_ResolveLookAtSub,up_in)),
-	LdSlot_ add_ui_self(R_TapePtr, S_(Binds_ResolveLookAtSub)),
+	LdSlot_ add_ui_self(    R_TapePtr, S_(Binds_ResolveLookAtSub)),
 
 	/* Stage up_in.x/y/z into the scratchpad. R_ScratchBase = R_SP = 0x1F800000. */
 	mac_load_v3s4( r.up_in, r.up_in_ptr, 0), LdSlot_
@@ -155,7 +155,7 @@ atom_info(atom_bind(Binds_ResolveLookAtSub)) MipsAtom_Proc_(aa, {
 })
 
 typedef Struct_(Binds_ResolveLookAt_PopulateMT3S4S2) {
-	U4 look_at;  /* MT3_S2S4* — destination matrix address */
+	MT3_S2S4* look_at;  /* MT3_S2S4* — destination matrix address */
 };
 typedef Struct_(RegUse_resolve_look_at_populate_mt3s4s2) {
 	Reg         look_at;
@@ -182,7 +182,7 @@ internal MipsAtom* AtomBundleEntry_(resolve_look_at,populate_mt3s4s2)(AtomArena_
 atom_info(atom_bind(Binds_ResolveLookAt_PopulateMT3S4S2)) MipsAtom_Proc_(aa, {
 	/* --- Tape pop: look_at pointer --- */
 	load_word(r.look_at, R_TapePtr, O_(Binds_ResolveLookAt_PopulateMT3S4S2,look_at)),
-	LdSlot_ add_ui_self(R_TapePtr, S_(Binds_ResolveLookAt_PopulateMT3S4S2)),
+	LdSlot_ add_ui_self( R_TapePtr, S_(Binds_ResolveLookAt_PopulateMT3S4S2)),
 
 	add_si(r.ux,  R_ScratchBase, O_(ResolveLookAtScratch, ux)), LdSlot_
 	add_si(r.uy,  R_ScratchBase, O_(ResolveLookAtScratch, uy)),
@@ -237,8 +237,8 @@ enum {
 };
 //screen_env_init. Mirrors the libpsyx's SetDefDispEnv + SetDefDrawEnv + the manual enable_auto_clear / initial_bg_color writes.
 internal MipsAtom_(screen_env_init) atom_info(atom_phase(screen_init)
-,	atom_reads(R_T0, R_ScreenX, R_ScreenY, R_ScreenBuf)
-, atom_writes(R_T0, R_ScreenX, R_ScreenY)
+,	atom_reads(R_ScreenBuf)
+, atom_writes(R_ScreenBuf)
 ) {
 	/* display[0] = (0, 0, 320, 240); rest of struct zeroed. */
 	add_ui(R_ScreenX, R_0, ScreenRes_X), add_ui(R_ScreenY, R_0, ScreenRes_Y),
@@ -288,23 +288,6 @@ internal MipsAtom_(screen_env_init) atom_info(atom_phase(screen_init)
 	mac_yield(),
 };
 
-/* gp_screen_init's GPR setup. Tests the mixed user-pinning + auto-reg pattern:
- *   - R_IO_BaseAddr = R_T4 (user-pinned via atom_reg; pre-existing)
- *   - R_GP1_Offset  = R_T2 (user-pinned via atom_reg; NEW -- for GPIO_PORT1_OFFSET)
- *   - R_ScreenX     = R_T5 (user-pinned via atom_reg; used as a transfer and GTE setup reg)
- *   - R_GpTmp       = auto-allocated by the lua pass and used for several GPU transfers;
- *                    the C preprocessor resolves it to the chosen free pool GPR.
- *
- * For gp_screen_init, the auto-reg pool exclusions are:
- *   user_pinned (from the corpus register_alias_registry) : R_T0..R_T7 (all 8 user-pinned across hello_camera.atom.c)
- *   body-parsed physical registers                         : aliases resolve through the registry;
- *                                                           the body uses R_ScreenX, not raw R_T5
- *   source_pool after both subtractions                   : {R_V0, R_V1} only
- *   R_GpTmp gets R_V0 (the first-fit choice). Its repeated GPU-transfer use proves that the
- *   auto-reg allocation is active while the R_ScreenX references prove the pinned alias is used.
- *   R_TapePtr (R_T9), R_AtomJmp (R_T8), R_AT are excluded from the POOL by construction in
- *   passes/auto_reg.lua -- see the "obvious exclusions" comment block at the top of that file.
- */
 enum {
 	R_IO_BaseAddr = R_T4 atom_reg, /* Caller-pinned: IO_BASE_ADDR = 0x1F800000 */
 	R_GP1_Offset  = R_T2 atom_reg, /* Caller-pinned: GPIO_PORT1_OFFSET = 0x10 */
@@ -458,7 +441,7 @@ typedef Struct_(Binds_PadInputCam) {
 	Camera*   cam;
 };
 internal MipsAtom_(pad_input_cam) atom_info(atom_bind(Binds_PadInputCam)
-, atom_reads( R_Cam, R_CamPadState, R_TapePtr)
+, atom_reads( R_Cam, R_CamPadState)
 , atom_writes(R_Cam)
 ) {
 	/* Bind pop: state → R_CamPadState (R_T5), cam → R_Cam (R_T4), advance R_TapePtr by 8. */
@@ -516,20 +499,17 @@ enum {
 #define R_OtBase_Code      R_T6_Code
 };
 typedef Struct_(Binds_CubeTri) {
-	U4     PrimCursor;
-	V4_S2* FaceCursor;
-	V3_S2* VertBase;
-	U4*    OtBase;
+	U1*    prim_cursor;
+	V4_S2* face_cursor;
+	V3_S2* vert_base;
+	U4*    ot_base;
 };
-internal MipsAtom_(rbind_cube_g4_face) atom_info(atom_bind(Binds_CubeTri), atom_phase(cube_g4)
-,	atom_reads(R_TapePtr)
-,	atom_writes(R_PrimCursor, R_FaceCursor, R_VertBase, R_OtBase, R_TapePtr)
-){
+internal MipsAtom_(rbind_cube_g4_face) atom_info(atom_bind(Binds_CubeTri), atom_phase(cube_g4)){
 	/* Pop 4 arguments from the tape directly into the workspace registers */
-	load_word(R_PrimCursor, R_TapePtr, O_(Binds_CubeTri,PrimCursor)),
-	load_word(R_FaceCursor, R_TapePtr, O_(Binds_CubeTri,FaceCursor)),
-	load_word(R_VertBase,   R_TapePtr, O_(Binds_CubeTri,VertBase)),
-	load_word(R_OtBase,     R_TapePtr, O_(Binds_CubeTri,OtBase)),
+	load_word(R_PrimCursor, R_TapePtr, O_(Binds_CubeTri,prim_cursor)),
+	load_word(R_FaceCursor, R_TapePtr, O_(Binds_CubeTri,face_cursor)),
+	load_word(R_VertBase,   R_TapePtr, O_(Binds_CubeTri,vert_base)),
+	load_word(R_OtBase,     R_TapePtr, O_(Binds_CubeTri,ot_base)),
 	LdSlot_ add_ui_self(    R_TapePtr, S_(Binds_CubeTri)),
 	mac_yield()
 };
@@ -545,11 +525,13 @@ MipsAtom_(cube_g4_face) atom_info(atom_phase(cube_g4),
 	load_half_u(R_T2, R_FaceCursor, 2 * S_(S2)),
 	// load_half_u(R_T3, R_FaceCursor, 3 * S_(S2)),
 
-	LdSlot_ mac_gte_load_tri_verts(R_VertBase, R_T0, R_T1, R_T2), GteDelay_ load_half_u(R_T3, R_FaceCursor, 3 * S_(S2)), LdSlot_
-	GteDelay_ nop, gte_cmdw_rotate_translate_perspective_triple,
+	LdSlot_ mac_gte_load_tri_verts(R_VertBase, R_T0, R_T1, R_T2), 
+		GteDelay_ load_half_u(R_T3, R_FaceCursor, 3 * S_(S2)), LdSlot_
+		GteDelay_ load_word(R_AtomJmp, R_TapePtr, 0),          LdSlot_ //ac_yield: word 1,
+	gte_cmdw_rotate_translate_perspective_triple,
 	gte_cmdw_nclip,
 
-	gte_mv_from_data_r(R_T0, C2_MAC0), GteDelay_ load_word(R_AtomJmp, R_TapePtr, 0), // ac_yield: word 1
+	gte_mv_from_data_r(R_T0, C2_MAC0), GteDelay_ add_ui_self(R_TapePtr, S_(MipsCode)), // ac_yield: word 2
 	branch_le_zero(R_T0, atom_offset(cull, cube_g4_face_exit)),
 		/* BD-slot: Write the prim tag (R_0=0; overwrites the legacy tag word in the prim_buffer).
 		 * If branch IS taken (face culled), the body is skipped and this 0-tag is stranded —
@@ -568,7 +550,7 @@ MipsAtom_(cube_g4_face) atom_info(atom_phase(cube_g4),
 		add_ui(      R_AT, R_0,  OrderingTbl_Len),
 		set_lt_u(    R_AT, R_T1, R_AT),
 
-		branch_equal(R_AT, R_0,  atom_offset(bounds_chk, cube_g4_face_exit)), BdSlot_ add_ui_self(R_TapePtr, S_(MipsCode)), // ac_yield: word 2
+		branch_equal(R_AT, R_0,  atom_offset(bounds_chk, cube_g4_face_exit)), BdSlot_ nop,
 			mac_insert_ot_tag(R_OtBase, R_PrimCursor, S_(Poly_G4)),
 			mac_format_g4_color(R_PrimCursor,
 				/* c0 magenta */ 0xFF, 0x00, 0xFF,
@@ -579,16 +561,16 @@ MipsAtom_(cube_g4_face) atom_info(atom_phase(cube_g4),
 // end: branch(cull)
 
 atom_label(cube_g4_face_exit)
-	add_ui_self(R_PrimCursor, S_(Poly_G4)),     /* 9 words = Poly_G4 */
-	add_ui_self(R_FaceCursor, S_(S2) * 4),      /* 4 × S2 = 8 bytes */
-	jump_reg(R_TapePtr), BdSlot_ nop // ac_yield: word 3-4
+	add_ui_self(R_PrimCursor, S_(Poly_G4)), // 9 words = Poly_G4
+	add_ui_self(R_FaceCursor, S_(S2) * 4),  // 4 × S2 = 8 bytes
+	jump_reg(R_AtomJmp), BdSlot_ nop      // ac_yield: word 3-4
 };
 
 typedef Struct_(Binds_FloorTri) {
-	U4     PrimCursor;
-	V3_S2* FaceCursor;
-	V3_S2* VertBase;
-	U4*    OtBase;
+	U1*    prim_cursor;
+	V3_S2* face_cursor;
+	V3_S2* vert_base;
+	U4*    ot_base;
 };
 internal
 MipsAtom_(rbind_floor_f3_face) atom_info(atom_bind(Binds_FloorTri), atom_phase(floor_f3)
@@ -596,10 +578,10 @@ MipsAtom_(rbind_floor_f3_face) atom_info(atom_bind(Binds_FloorTri), atom_phase(f
 	, atom_writes(R_PrimCursor, R_FaceCursor, R_VertBase, R_OtBase, R_TapePtr)
 ){
 	/* Pop 4 arguments from the tape directly into the workspace registers */
-	load_word(R_PrimCursor, R_TapePtr, O_(Binds_FloorTri,PrimCursor)),
-	load_word(R_FaceCursor, R_TapePtr, O_(Binds_FloorTri,FaceCursor)),
-	load_word(R_VertBase,   R_TapePtr, O_(Binds_FloorTri,VertBase)),
-	load_word(R_OtBase,     R_TapePtr, O_(Binds_FloorTri,OtBase)),
+	load_word(R_PrimCursor, R_TapePtr, O_(Binds_FloorTri,prim_cursor)),
+	load_word(R_FaceCursor, R_TapePtr, O_(Binds_FloorTri,face_cursor)),
+	load_word(R_VertBase,   R_TapePtr, O_(Binds_FloorTri,vert_base)),
+	load_word(R_OtBase,     R_TapePtr, O_(Binds_FloorTri,ot_base)),
 	LdSlot_ add_ui_self(    R_TapePtr, S_(Binds_FloorTri)),
 	mac_yield()
 };
@@ -638,13 +620,12 @@ MipsAtom_(floor_f3_face) atom_info(atom_phase(floor_f3)
 /* Advance Input Cursor & Yield (Both branch targets land here) */
 atom_label(floor_f3_face_exit)
 	add_ui_self(R_FaceCursor, S_(S2) * 4),  /* Advance Face Cursor (4 * S2 = 8 bytes) */
-	jump_reg(R_TapePtr), BdSlot_ nop // ac_yield: word 3-4
+	jump_reg(R_AtomJmp), BdSlot_ nop // ac_yield: word 3-4
 };
 
-typedef Struct_(Binds_SyncPrimitiveArena) { U4 used; U4 cursor; };
+typedef Struct_(Binds_SyncPrimitiveArena) { U4* used; U1* cursor; };
 internal MipsAtom_(sync_primitive_arena) atom_info(atom_bind(Binds_SyncPrimitiveArena)
-	, atom_reads( R_TapePtr, R_PrimCursor)
-	, atom_writes(R_TapePtr)
+	, atom_reads(R_PrimCursor), atom_writes(R_AT)
 ){
 	load_word(R_AT, R_TapePtr, O_(Binds_SyncPrimitiveArena,used)),
 	load_word(R_T0, R_TapePtr, O_(Binds_SyncPrimitiveArena,cursor)), LdSlot_

code/hello_camera/hello_camera.c

@@ -24,7 +24,7 @@
 #include "duffle/pad.h"
 
 #include "duffle/dsl.atom.h"
-#include "duffle/lottes_tape.h"
+#include "duffle/tape.h"
 
 #include "duffle/bios.h"
 #include "duffle/psyq.h"
@@ -80,9 +80,6 @@ typedef Struct_(SMemory) {
 	PadBiosRaw pad_raw[2];
 	PadState   pad[2];
 
-	// TODO(Ed): We don't need this we can just cast at any point an address to a desired view of scratchpad, we have the address.
-	U4_V scratchpad; // d-cache
-
 	U1 ct_init_atom_mem[CT_InitAtomMem_Size];
 	MipsAtom* normalize_v3s4;
 	MipsAtom* gte_cross_v3s4;
@@ -96,11 +93,11 @@ extern SMemory smem;
 #define pad0_btn_(btn) btn & smem.pad[0].buttons
 #define pad1_btn_(btn) btn & smem.pad[1].buttons
 
-I_ B1* prim__alloc(U4 type_width, Str8 type_name) {
+I_ U1* prim__alloc(U4 type_width, Str8 type_name) {
 	gknown PrimitiveArena* pa  = & smem.primitives;
-	gknown B1*             buf = (B1*) r_(smem.primitives.buf)[smem.active_buf_id];
+	gknown U1*             buf = (U1*) r_(smem.primitives.buf)[smem.active_buf_id];
 	assert(pa->used + type_width < PrimitiveBuff_Len);
-	B1* next  = buf + pa->used;
+	U1* next  = buf + pa->used;
 	pa->used += type_width;
 	return next;
 }
@@ -114,10 +111,10 @@ I_ void resolve_look_at_c11(MT3_S2S4* look_at, P3_S4* eye, P3_S4* target, V3_S4*
 	V3_S4 pos, off;
 
 	forward = target[0]; sub_v3s4(& forward, eye[0]); // RGA(Lengyel): Affine point - point = zero-weight direction.
-	normalize_v3s4(& forward, & uz);                  // RGA(Lengyel): Normalize the direction bulk. Not finite-point unitization.
+	psy_normalize_v3s4(& forward, & uz);              // RGA(Lengyel): Normalize the direction bulk. Not finite-point unitization.
 
-	cross_v3s4(& uz,   up_in, & right); normalize_v3s4(& right, & ux); // RGA(Lengyel): Complement(Wedge(forward, up_in)) -> right axis.
-	cross_v3s4(& uz, & ux,    & up);    normalize_v3s4(& up,    & uy); // RGA(Lengyel): Complement(Wedge(forward, right)) -> up axis.
+	cross_v3s4(& uz,   up_in, & right); psy_normalize_v3s4(& right, & ux); // RGA(Lengyel): Complement(Wedge(forward, up_in)) -> right axis.
+	cross_v3s4(& uz, & ux,    & up);    psy_normalize_v3s4(& up,    & uy); // RGA(Lengyel): Complement(Wedge(forward, right)) -> up axis.
 
 	// RGA(Lengyel): matrix expansion of the world-to-camera rotation (basis rows).
 	look_at->m[0][0] = ux.x; look_at->m[0][1] = ux.y; look_at->m[0][2] = ux.z;
@@ -141,16 +138,16 @@ internal void compile_init_atoms(void) {
 
 	smem.gte_cross_v3s4 = gte_cross_v3s4(& ab,
 		RegUse_(gte_cross_v3s4) {
-			.a  = ralloc_v3(),
-			.b  = ralloc_v3(),
-			.x  = ralloc(),
-			.y  = ralloc(),
-			.z  = ralloc(),
+			.a     = ralloc_v3(),
+			.b     = ralloc_v3(),
+			.out   = ralloc(),
+			.src_a = ralloc(),
+			.src_b = ralloc(),
 		});
 	regfile_reset(& rf);
 
-	smem.normalize_v3s4 = build_normalize_v3s4(& ab,
-		RegUse_(build_normalize_v3s4) {
+	smem.normalize_v3s4 = normalize_v3s4(& ab,
+		RegUse_(normalize_v3s4) {
 			.res = ralloc_v3(),
 			.r0     = ralloc(),
 			.r1     = ralloc(),
@@ -167,15 +164,11 @@ internal void compile_init_atoms(void) {
 }
 
 internal void compile_resolve_look_at(void) {
-	AtomArena ab = atomarena_make(slice_ut_arr(smem.resolve_look_at_mem));
 	AtomBundle_resolve_look_at_R bundle = C_(void*, smem.resolve_look_at_bundle);
-
-	/* R_ScratchBase (= R_SP) is a tape carrier preserved across atoms; no carrier
-	 * pin is needed in the regfile. The standard 24-register pool is sufficient. */
-	RegFile rf = regfile(regfile_abi_mask);
+	AtomArena ab = atomarena_make(slice_ut_arr(smem.resolve_look_at_mem));
+	RegFile   rf = regfile(regfile_abi_mask);
 #define ralloc() regfile_alloc(& rf)
 #define ralloc_v3() { ralloc(), ralloc(), ralloc() }
-
 	bundle->input_and_sub = AtomBundleEntry_(resolve_look_at, input_and_sub)(& ab,
 		RegUse_(resolve_look_at_input_and_sub) {
 			.target_ptr = ralloc(),
@@ -203,78 +196,63 @@ internal void compile_resolve_look_at(void) {
 			.r2      = ralloc(),
 		});
 
-	/* Sanity check: arena didn't overflow. */
-	assert(ab.used <= ResolveLookAtArena_Size);
+	assert(ab.used <= ResolveLookAtArena_Size); // Sanity check: arena didn't overflow.
 #undef ralloc
 }
 
-/* Emit the resolve_look_at bundle into the tape. Called once per frame from update(). */
-I_ void resolve_look_at(TapeBuilder_R tb
-	,	MT3_S2S4* look_at
-	,	P3_S4*    eye
-	,	P3_S4*    target
-	,	V3_S4*    up_in
-){
+// Emit the resolve_look_at bundle into the tape. Called once per frame from update().
+I_ void resolve_look_at(TapeBuilder_R tb,	MT3_S2S4* look_at, P3_S4* eye, P3_S4* target,	V3_S4* up_in) {
 	/* Typed view of the scratchpad for field-address arithmetic. */
-	ResolveLookAtScratch* sp = C_scratch(ResolveLookAtScratch*);
+	ResolveLookAtScratch*        sp     = C_scratch(ResolveLookAtScratch*);
 	AtomBundle_resolve_look_at_R bundle = C_(void*, smem.resolve_look_at_bundle);
-
-	tb_emit(tb, bundle->input_and_sub); {
-		tb_data(tb, u4_(target));
-		tb_data(tb, u4_(eye));
-		tb_data(tb, u4_(up_in));
-	}
-	tb_emit(tb, bundle->normalize_fwd_uz); {
-		tb_data(tb, u4_(O_(ResolveLookAtScratch, fwd) | (O_(ResolveLookAtScratch, uz) << 16)));
-	}
-	tb_emit(tb, bundle->cross_to_right); {
-		tb_data(tb, u4_(& sp->uz));
-		tb_data(tb, u4_(& sp->up_in));
-		tb_data(tb, u4_(& sp->right));
-	}
-	tb_emit(tb, bundle->normalize_right_ux); {
-		tb_data(tb, u4_(O_(ResolveLookAtScratch, right) | (O_(ResolveLookAtScratch, ux) << 16)));
-	}
-	tb_emit(tb, bundle->cross_to_up); {
-		tb_data(tb, u4_(& sp->uz));
-		tb_data(tb, u4_(& sp->ux));
-		tb_data(tb, u4_(& sp->up));
-	}
-	tb_emit(tb, bundle->normalize_up_uy); {
-		tb_data(tb, u4_(O_(ResolveLookAtScratch, up) | (O_(ResolveLookAtScratch, uy) << 16)));
-	}
-	tb_emit(tb, bundle->populate_mt3s4s2); {
-		tb_data(tb, u4_(look_at));
-	}
+	tb_emit(tb, bundle->input_and_sub); tb_bind_(tb, Binds_ResolveLookAtSub, 
+		.target = target,
+		.eye    = eye,
+		.up_in  = up_in,
+	);
+	tb_emit(tb, bundle->normalize_fwd_uz); tb_bind_(tb, Binds_normalize_v3s4,
+		.src_offset = O_(ResolveLookAtScratch,fwd),
+		.dst_offset = O_(ResolveLookAtScratch,uz),
+	);
+	tb_emit(tb, bundle->cross_to_right); tb_bind_(tb, Binds_gte_cross_v3s4,
+		.src_a = & sp->uz,
+		.src_b = & sp->up_in,
+		.out   = & sp->right,
+	);
+	tb_emit(tb, bundle->normalize_right_ux); tb_bind_(tb, Binds_normalize_v3s4, 
+		.src_offset = O_(ResolveLookAtScratch,right),
+		.dst_offset = O_(ResolveLookAtScratch,ux),
+	);
+	tb_emit(tb, bundle->cross_to_up); tb_bind_(tb, Binds_gte_cross_v3s4, 
+		.src_a = & sp->uz,
+		.src_b = & sp->ux,
+		.out   = & sp->up,
+	);
+	tb_emit(tb, bundle->normalize_up_uy); tb_bind_(tb, Binds_normalize_v3s4,
+		.src_offset = O_(ResolveLookAtScratch,up),
+		.dst_offset = O_(ResolveLookAtScratch,uy),
+	);
+	tb_emit(tb, bundle->populate_mt3s4s2); tb_bind_(tb, Binds_ResolveLookAt_PopulateMT3S4S2, 
+		.look_at = look_at,
+	);
 }
+FI_ void camera_look_at(TapeBuilder_R tb, Camera* c, P3_S4* target, V3_S4* up_in) { resolve_look_at(tb, & c->look_at, & c->pos, target, up_in); }
 
-GCC_OPTIMIZATION_DISABLE
 void update(PrimitiveArena* pa, U4* ordering_buf) 
 {
 	TapeBuilder tb = tb_make(slice_ut_arr(smem.MemTape));
-
-	// Pad Input
-	{
+	/*Pad Input*/ {
 		tb.used = 0; tb_scope_run(& tb) {
 			// Grab latest state from bios.
 			tb_emit_(pad_bios_snapshot);
-				tb_data_(raw,   & smem.pad_raw[0]);
-				tb_data_(state, & smem.pad[0]);
-			// tb_emit_(pad_bios_snapshot);
-			// 	tb_data_(raw,   & smem.pad_raw[1]);
-			// 	tb_data_(state, & smem.pad[1]);
-
+				tb_data(& tb, u4_(& smem.pad_raw[0]));
+				tb_data(& tb, u4_(& smem.pad[0]));
 			tb_emit_(pad_input_cam);
-				tb_data_(state, & smem.pad[0]);
-				tb_data_(cam, & smem.cam);
-
-			// tb_emit_(pad_input_cube_rotation);
-			// 	tb_data_(state,     & smem.pad[0]);
-			// 	tb_data_(cube_rot,  & smem.cube.rot);
-			// 	tb_data_(floor_rot, & smem.floor.rot);
+				tb_data(& tb, u4_(& smem.pad[0]));
+				tb_data(& tb, u4_(& smem.cam));
 		}
 	}
-
+	
 	orderingtbl_clear_reverse(ordering_buf, OrderingTbl_Len);
 
 	// Update the position based on acceleration and velocity
@@ -299,7 +277,7 @@ void update(PrimitiveArena* pa, U4* ordering_buf)
 	if (use_c11_path == false)
 	{
 		tb.used = 0; tb_scope_run(& tb) {
-			resolve_look_at(& tb, & smem.cam.look_at, & smem.cam.pos, & smem.cube.pos, & v3s4(0, -fp_one, 0));
+			camera_look_at(& tb, & smem.cam, & smem.cube.pos, & v3s4(0, -fp_one, 0));
 		}
 	}
 
@@ -309,33 +287,30 @@ void update(PrimitiveArena* pa, U4* ordering_buf)
 		mt3s2s4_rotation   (& smem.cube.rot,    & smem.tform_world);
 		mt3s2s4_translation(& smem.tform_world, & smem.cube.pos);
 		mt3s2s4_scale      (& smem.tform_world, & smem.cube.scale);
-
 		// Combine world and look_at matrix.
 		gte_comp_coord_m3s2(& smem.cam.look_at, & smem.tform_world, & smem.tform_view);
 		gte_matrix_set_rotation   (& smem.tform_view);
 		gte_matrix_set_translation(& smem.tform_view);
 
-		U4 prim_base   = u4_(pa->buf[smem.active_buf_id]);
-		U4 prim_cursor = prim_base + pa->used;
-
-		tb.used = 0; tb_scope(& tb) {
-			tb_emit(& tb, rbind_cube_g4_face);
-				tb_data(& tb, prim_cursor);
-				tb_data(& tb, u4_(smem.cube.faces));
-				tb_data(& tb, u4_(smem.cube.verts));
-				tb_data(& tb, u4_(ordering_buf));
-
+		// TODO(Ed): We should do a bounds check beforehand to confirm pa can hold all tris?
+		// The tape atoms in-flight should not need to care.
+		U1* prim_base   = u1_r(pa->buf[smem.active_buf_id]);
+		U1* prim_cursor = prim_base + pa->used;
+		tb.used = 0; tb_scope_run(& tb) {
+			tb_emit(& tb, rbind_cube_g4_face); tb_bind_(& tb, Binds_CubeTri, 
+				.prim_cursor = prim_cursor,
+				.face_cursor = smem.cube.faces,
+				.vert_base   = smem.cube.verts,
+				.ot_base     = ordering_buf,
+			);
 			for (U4 i = 0; i < Cube_num_faces; i++) {
-				// Two triangles per quad face: (x,y,z) and (x,z,w)
-				tb_emit(& tb, cube_g4_face);
+				tb_emit(& tb, cube_g4_face); // Two triangles per quad face: (x,y,z) and (x,z,w)
 			}
-
-			tb_emit(& tb, sync_primitive_arena);
-				tb_data(& tb, u4_(& pa->used));
-				tb_data(& tb, prim_base);
+			tb_emit(& tb, sync_primitive_arena); tb_bind_(& tb, Binds_SyncPrimitiveArena,
+				.used   = & pa->used,
+				.cursor = prim_base,
+			);
 		}
-		tape_run(tb_slice(tb));// Fire off the tape (bigger-clobber variant).
-
 		// smem.cube.rot.y += 30;
 	}
 	// Draw floor
@@ -344,45 +319,35 @@ void update(PrimitiveArena* pa, U4* ordering_buf)
 		mt3s2s4_rotation   (& smem.floor.rot,   & smem.tform_world);
 		mt3s2s4_translation(& smem.tform_world, & smem.floor.pos);
 		mt3s2s4_scale      (& smem.tform_world, & smem.floor.scale);
-
 		// Combine world and look_at matrix.
 		gte_comp_coord_m3s2(& smem.cam.look_at, & smem.tform_world, & smem.tform_view);
-
 		gte_matrix_set_rotation   (& smem.tform_view);
 		gte_matrix_set_translation(& smem.tform_view);
 
-		U4 prim_base   = u4_(pa->buf[smem.active_buf_id]);
-		U4 prim_cursor = prim_base + pa->used;
-
 		// TODO(Ed): We should do a bounds check beforehand to confirm pa can hold all tris?
 		// The tape atoms in-flight should not need to care.
-
-		// Prepare the tape. (Push protocol to tape)
-		tb.used = 0; tb_scope(& tb) {
-			// tb_emit(& tb, set_gte_mt3s2s4);
-			// 	tb_data(& tb, u4_(& smem.tform_view));
-
-			tb_emit(& tb, rbind_floor_f3_face);
-			// TODO(Ed): Just use a single context struct ref?
-				tb_data(& tb, prim_cursor);
-				tb_data(& tb, u4_(smem.floor.faces));
-				tb_data(& tb, u4_(smem.floor.verts));
-				tb_data(& tb, u4_(ordering_buf));
+		U1_R prim_base   = u1_r(pa->buf[smem.active_buf_id]);
+		U1_R prim_cursor = prim_base + pa->used;
+		tb.used = 0; tb_scope_run(& tb) { // Prepare the tape. (Push protocol to tape)
+			tb_emit(& tb, rbind_floor_f3_face); tb_bind_(& tb, Binds_FloorTri,
+				.prim_cursor = prim_cursor,
+				.face_cursor = smem.floor.faces,
+				.vert_base   = smem.floor.verts,
+				.ot_base     = ordering_buf,
+			);
 			for (U4 i = 0; i < Floor_num_faces; i++) {
 				tb_emit(& tb, floor_f3_face);
 			}
 			// After floor_f3_face iterations complete, the primitive arena's used counter needs updating.
-			tb_emit(& tb, sync_primitive_arena);
-				tb_data(& tb, u4_(& pa->used));
-				tb_data(& tb, prim_base);
+			tb_emit(& tb, sync_primitive_arena); tb_bind_(& tb, Binds_SyncPrimitiveArena,
+				.used   = & pa->used,
+				.cursor = prim_base,
+			);
 		}
-		tape_run(tb_slice(tb));// Fire off the tape (bigger-clobber variant).
-
 		// C-side state (pa->used) has already been updated by the tape!
 		// smem.floor.rot.y += 5;
 	}
 }
-GCC_OPTIMIZATION_ENABLE
 
 void render(void) {
 }
@@ -399,12 +364,22 @@ void gp_display_frame(DoubleBuffer* screen_buf, S4* active_buf_id, U4* ordering_
 	active_buf_id[0] = ! active_buf_id[0]; // Swap current buffer
 }
 
-GCC_OPTIMIZATION_DISABLE
 int main(void)
 {
 	smem = (SMemory){0};
-	// TODO(Ed): remove this field we don't need it in smem.
-	smem.scratchpad = C_(U4_V, Scratchpad_Loc);
+
+	B4 basic_sample = false; if (basic_sample) {
+		// We will be defining the tape here along with its atom, then running the tape after.
+		TapeBuilder tb = tb_make(slice_ut_arr(smem.MemTape));
+		MipsCode add_one_to_R_T1[] = {
+			add_ui_self(R_T1, 1),
+			mac_yield(),
+		};
+		tb_emit(& tb, C_(MipsAtom*, add_one_to_R_T1));
+		Tape tape = tb_end(& tb);
+		tape_run(tape);
+	}
+
 	// smem.primitives.used = 0;
 	// smem.active_buf_id   = 0;
 	smem.cam.pos = v3s4(500, -1000, -1500);
@@ -449,4 +424,3 @@ int main(void)
 	};
 	return 0;
 }
-GCC_OPTIMIZATION_ENABLE

code/hello_camera/hello_camera.h

@@ -24,7 +24,7 @@ enum {
 typedef U4 OrderingTable_Buffer[OrderingTbl_Len];
 typedef Array_(OrderingTable_Buffer, 2);
 
-typedef B1 PrimitiveBuffer[PrimitiveBuff_Len];
+typedef U1 PrimitiveBuffer[PrimitiveBuff_Len];
 typedef Array_(PrimitiveBuffer, 2);
 typedef Struct_(PrimitiveArena) {
 	A2_PrimitiveBuffer buf;
@@ -54,14 +54,14 @@ I_ void ent_cube128_init(A8_V3_S2* verts, A6_V4_S2* faces) {
 		{ 2, 3, 6, 7 },
 		{ 3, 0, 7, 4 },
 	};
-	mem_copy(u4_(verts), u4_(& baked_verts), S_(A8_V3_S2) );
-	mem_copy(u4_(faces), u4_(& baked_faces), S_(A6_V4_S2) );
+	mem_copy(b1_r(verts), b1_r(& baked_verts), S_(A8_V3_S2) );
+	mem_copy(b1_r(faces), b1_r(& baked_faces), S_(A6_V4_S2) );
 	return;
 }
 typedef Struct_(Ent_Cube) {
 	V3_S4 accel;
 	V3_S4 vel;
-	V3_S4 pos; // RGA(Lengyel): affine point with implicit weight one. Storage alias of V3_S4.
+	V3_S4 pos;
 	V3_S4 scale;
 	V3_S2 rot;
 	A8_V3_S2 verts;
@@ -83,12 +83,12 @@ I_ void ent_floor_init(A4_V3_S2* verts, A2_V3_S2* faces) {
 		{ 0, 1, 2 },
 		{ 1, 3, 2 },
 	};
-	mem_copy(u4_(verts), u4_(& baked_verts), S_(A4_V3_S2));
-	mem_copy(u4_(faces), u4_(& baked_faces), S_(A2_V3_S2));
+	mem_copy(b1_r(verts), b1_r(& baked_verts), S_(A4_V3_S2));
+	mem_copy(b1_r(faces), b1_r(& baked_faces), S_(A2_V3_S2));
 };
 typedef Struct_(Ent_Floor) {
 	V3_S4 accel;
-	V3_S4 pos; // RGA(Lengyel): affine point with implicit weight one. Storage alias of V3_S4.
+	V3_S4 pos;
 	V3_S4 scale;
 	V3_S2 rot;
 	A4_V3_S2 verts;
@@ -96,7 +96,7 @@ typedef Struct_(Ent_Floor) {
 };
 
 typedef Struct_(Camera) {
-	P3_S4    pos; // RGA(Lengyel): affine point with implicit weight one. Storage alias of V3_S4.
+	P3_S4    pos;
 	V3_S2    rot;
 	MT3_S2S4 look_at;
 };

scripts/duffle_emit.lua

@@ -1041,6 +1041,16 @@ function M.find_atom_proc_decl_for(source, before_pos, mips_atom_ptr_len)
 		if source:sub(next_pos, next_pos) == "(" then
 			local inner, after_paren = M.read_parens(source, next_pos) ---@type string|nil, integer
 			if inner then
+				if ident == "AtomBundleEntry_" then
+					local tmpl = M.split_top_level_commas(inner) ---@type string[]
+					if   #tmpl ~= 2 then return nil end
+					local name        = M.trim(tmpl[1]) .. "_" .. M.trim(tmpl[2]) ---@type string
+					local formals_pos = M.skip_ws_and_cmt(source, after_paren)    ---@type integer
+					if source:sub(formals_pos, formals_pos) ~= "(" then return nil end
+					local  real_inner, after_real = M.read_parens(source, formals_pos) ---@type string|nil, integer
+					if not real_inner then return nil end
+					return name, real_inner, name, after_real
+				end
 				return ident, inner, ident, after_paren
 			end
 		end

scripts/passes/report.lua

@@ -544,6 +544,50 @@ local function render_section_binds(add, view)
 	if not wrote then add("_(none)_"); add("") end
 end
 
+--- @param add  fun(s: string): nil
+--- @param view ModuleView
+--- @return nil
+local function render_section_atom_bundles(add, view)
+	local bundles = (view.corpus and view.corpus.atom_bundles) or {} ---@type table<string, AtomBundle>
+	local names   = {}                                               ---@type string[]
+	for name, bundle in pairs(bundles) do ---@type string, AtomBundle
+		if path_in_module(bundle.path, view) then
+			names[#names + 1] = name
+		end
+	end
+	if #names == 0 then add("_(none)_"); add(""); return end
+	table.sort(names)
+	for _, name in ipairs(names) do ---@type integer, string
+		local bundle  = bundles[name]        ---@type AtomBundle
+		local entries = bundle.entries or {} ---@type table<string, string>
+		add(string.format("### %s", name))
+		for _, slot in ipairs(bundle.slots or {}) do ---@type integer, string
+			local ident = entries[slot] ---@type string|nil
+			if ident then
+				add(string.format("- `%s` `%s`", slot, ident))
+			else
+				add(string.format("- `%s`", slot))
+			end
+		end
+		add("")
+	end
+end
+
+--- @param add  fun(s: string): nil
+--- @param view ModuleView
+--- @return nil
+local function render_section_tape_emits(add, view)
+	local wrote = false ---@type boolean
+	for _, src in ipairs(view.sources or {}) do ---@type integer, SourceFile
+		for _, emit in ipairs((src.scan and src.scan.tape_emits) or {}) do ---@type integer, TapeEmit
+			wrote = true
+			add(string.format("- `%s` `%s`", emit.name or "?", emit.binds or "—"))
+		end
+	end
+	if not wrote then add("_(none)_"); add(""); return end
+	add("")
+end
+
 --- @param add fun(s: string): nil
 --- @param view ModuleView
 --- @return nil
@@ -895,6 +939,8 @@ local SECTION_RENDERERS = { ---@type SectionRenderer[]
 	{ header = "## Annotations",           render = render_section_annotations },
 	{ header = "## Component annotations", render = render_section_component_annotations },
 	{ header = "## Binds_* structs",       render = render_section_binds },
+	{ header = "## Atom bundles",          render = render_section_atom_bundles },
+	{ header = "## Tape emits",            render = render_section_tape_emits },
 	{ header = "## Phases / views / ctx",  render = render_section_phases },
 	{ header = "## Register aliases",      render = render_section_aliases },
 	{ header = "## Auto-reg",              render = render_section_autoreg },

scripts/passes/scan_source.lua

@@ -127,6 +127,13 @@ local parse_enum_int_literal ---@type fun(text: string, start: integer): (intege
 --- @field body   string
 --- @field bytes  integer|nil
 
+--- @class AtomBundle
+--- @field name    string
+--- @field slots   string[]   -- typedef order
+--- @field line    integer
+--- @field path    string
+--- @field entries table<string, string>|nil  -- slot → B_E when an AtomBundleEntry_ proc exists
+
 --- @class AliasEntry
 --- @field name          string
 --- @field code          integer
@@ -182,6 +189,14 @@ local parse_enum_int_literal ---@type fun(text: string, start: integer): (intege
 --- @class TapeChain
 --- @field [integer] string  -- ordered atom names in one tb_emit chain
 
+--- @class TapeEmit
+--- @field name       string
+--- @field binds      string|nil
+--- @field line       integer
+--- @field path       string
+--- @field slot       string|nil
+--- @field data_words integer|nil
+
 --- @class RegUseView
 --- @field names string[]
 --- @field lanes boolean
@@ -208,6 +223,8 @@ local parse_enum_int_literal ---@type fun(text: string, start: integer): (intege
 --- @field reg_use_schemas         table<string, RegUseSchema>
 --- @field reg_use_errors          RegUseError[]
 --- @field tape_chains             TapeChain[]
+--- @field tape_emits              TapeEmit[]
+--- @field atom_bundles            table<string, AtomBundle>
 --- @field _source_file            string|nil
 --- @field _code_macros            table<string, integer>|nil  -- bag
 --- @field _code_macro_bodies      table<string, string>|nil  -- bag
@@ -2439,11 +2456,76 @@ local function parse_typedef_array(source, pos, id2_end, line_of, out, after_typ
 	return semi and (semi + 1) or after_paren
 end
 
+--- Parse `MipsAtom *slot, …;` from an AtomBundle_ typedef body.
+--- Ignores the MipsAtom type token. Slot name is the ident after `*`.
+--- @param body string
+--- @return string[]
+local function parse_atom_bundle_slots(body)
+	local slots = {} ---@type string[]
+	local pos   = 1  ---@type integer
+	while pos <= #body do
+		pos = duffle.skip_ws_and_cmt(body, pos)
+		if pos > #body then break end
+		local b = body:byte(pos) ---@type integer
+		if b == BYTE_SEMI then
+			break
+		elseif b == BYTE_COMMA then
+			pos = pos + 1
+		elseif b == BYTE_STAR then
+			local after_star       = duffle.skip_ws_and_cmt(body, pos + 1) ---@type integer
+			local ident, ident_end = duffle.read_ident(body, after_star)   ---@type string|nil, integer
+			if ident then
+				slots[#slots + 1] = ident
+				pos = ident_end
+			else
+				pos = pos + 1
+			end
+		else
+			local ident, ident_end = duffle.read_ident(body, pos) ---@type string|nil, integer
+			if ident then
+				pos = ident_end
+			else
+				pos = pos + 1
+			end
+		end
+	end
+	return slots
+end
+
+-- Shape 5: `typedef AtomBundle_(<name>) { MipsAtom *slot, … };`
+--- @param source string
+--- @param pos integer
+--- @param id2_end integer
+--- @param line_of fun(pos: integer): integer
+--- @param out SourceScan
+--- @param after_typedef integer
+--- @return integer
+local function parse_typedef_atom_bundle(source, pos, id2_end, line_of, out, after_typedef)
+	local inner, after_paren, open_paren = read_parens_after(source, id2_end, id2_end) ---@type string|nil, integer, integer
+	if not inner then return id2_end end
+	local name = duffle.trim(inner) ---@type string
+
+	local body, after_brace = find_body_braces(source, after_paren, open_paren + 1) ---@type string|nil, integer
+	if not body then return after_brace end
+	if name ~= "" and out.atom_bundles[name] == nil then
+		local bundle = { ---@type AtomBundle
+			name  = name,
+			slots = parse_atom_bundle_slots(body),
+			line  = line_of(pos),
+			path  = out._source_file or "",
+		}
+		out.atom_bundles[name] = bundle
+	end
+	attach_debug_skip_marker(out, "unrelated")
+	return after_brace
+end
+
 local TYPE_FORMS = { ---@type table<string, fun(source: string, pos: integer, id2_end: integer, line_of: fun(pos: integer): integer, out: SourceScan, after_typedef: integer): integer>
-	Struct_ = parse_typedef_struct,
-	Enum_   = parse_typedef_enum,
-	TSet_   = parse_typedef_tset,
-	Array_  = parse_typedef_array,
+	Struct_     = parse_typedef_struct,
+	Enum_       = parse_typedef_enum,
+	TSet_       = parse_typedef_tset,
+	Array_      = parse_typedef_array,
+	AtomBundle_ = parse_typedef_atom_bundle,
 }
 
 --- Parse: `typedef` declarations.
@@ -2827,6 +2909,23 @@ local function parse_addrs_assign(source, pos, ident_end, line_of, out)
 	return rhs
 end
 
+--- @param out   SourceScan
+--- @param name  string
+--- @param last  string
+--- @param line  integer
+--- @return nil
+local function push_tape_emit(out, name, last, line)
+	out.tape_emits = out.tape_emits or {}
+	out.tape_emits[#out.tape_emits + 1] = { ---@type TapeEmit
+		name       = name,
+		binds      = nil,
+		line       = line,
+		path       = out._source_file or "",
+		slot       = last:find("->", 1, true) and name or nil,
+		data_words = 0,
+	}
+end
+
 --- @param source string
 --- @param pos integer
 --- @param ident_end integer
@@ -2837,10 +2936,12 @@ local function parse_tb_emit_(source, pos, ident_end, line_of, out)
 	local after = duffle.skip_ws_and_cmt(source, ident_end) ---@type integer
 	if source:sub(after, after) ~= "(" then return ident_end end
 	local inner, after_p = duffle.read_parens(source, after)           ---@type string|nil, integer
-	local name           = duffle.trim(inner or ""):match("^([%w_]+)") ---@type string
+	local last           = duffle.trim(inner or "")                    ---@type string
+	local name           = last:match("^([%w_]+)")                     ---@type string
 	if name then
 		out._chain = out._chain or {}
 		out._chain[#out._chain + 1] = name
+		push_tape_emit(out, name, last, line_of(pos))
 	end
 	return after_p or (after + 1)
 end
@@ -2865,6 +2966,49 @@ local function parse_tb_emit(source, pos, ident_end, line_of, out)
 	if name then
 		out._chain = out._chain or {}
 		out._chain[#out._chain + 1] = name
+		push_tape_emit(out, name, last, line_of(pos))
+	end
+	return after_p or (after + 1)
+end
+
+--- @param source string
+--- @param pos integer
+--- @param ident_end integer
+--- @param line_of fun(pos: integer): integer
+--- @param out SourceScan
+--- @return integer
+local function parse_tb_bind_(source, pos, ident_end, line_of, out)
+	local after = duffle.skip_ws_and_cmt(source, ident_end) ---@type integer
+	if source:sub(after, after) ~= "(" then return ident_end end
+	local inner, after_p = duffle.read_parens(source, after)          ---@type string|nil, integer
+	local args           = duffle.split_top_level_commas(inner or "") ---@type string[]
+	local typ            = duffle.trim(args[2] or "")                 ---@type string
+	if typ ~= "" then
+		local emits = out.tape_emits or {} ---@type TapeEmit[]
+		for i = #emits, 1, -1 do           ---@type integer
+			if emits[i].binds == nil then
+				emits[i].binds = typ
+				break
+			end
+		end
+	end
+	return after_p or (after + 1)
+end
+
+--- @param source string
+--- @param pos integer
+--- @param ident_end integer
+--- @param line_of fun(pos: integer): integer
+--- @param out SourceScan
+--- @return integer
+local function parse_tb_data(source, pos, ident_end, line_of, out)
+	local after = duffle.skip_ws_and_cmt(source, ident_end) ---@type integer
+	if source:sub(after, after) ~= "(" then return ident_end end
+	local _, after_p = duffle.read_parens(source, after) ---@type string|nil, integer
+	local emits = out.tape_emits or {} ---@type TapeEmit[]
+	local last  = emits[#emits]        ---@type TapeEmit|nil
+	if last then
+		last.data_words = (last.data_words or 0) + 1
 	end
 	return after_p or (after + 1)
 end
@@ -2872,6 +3016,8 @@ end
 local C_STMT_PARSERS = { ---@type table<string, fun(source: string, pos: integer, ident_end: integer, line_of: fun(pos: integer): integer, out: SourceScan): integer>
 	tb_emit_ = parse_tb_emit_,
 	tb_emit  = parse_tb_emit,
+	tb_bind_ = parse_tb_bind_,
+	tb_data  = parse_tb_data,
 	addrs    = parse_addrs_assign,
 }
 
@@ -2924,6 +3070,8 @@ local function scan_source(source, source_file, code_macros, code_macro_bodies)
 		atoms                = {},
 		raw_atoms            = {},
 		binds                = {},
+		atom_bundles         = {},
+		tape_emits           = {},
 		atom_infos           = {},
 		component_atom_infos = {},
 		macros               = {},
@@ -3238,6 +3386,8 @@ local function merge_corpus_registries(corpus)
 	corpus.reg_use_schemas         = corpus.reg_use_schemas         or {}
 	corpus.reg_use_errors          = corpus.reg_use_errors          or {}
 	corpus.tape_chains             = corpus.tape_chains             or {}
+	corpus.atom_bundles            = corpus.atom_bundles            or {}
+	corpus.tape_emits              = corpus.tape_emits              or {}
 
 	-- Replace the existing corpus collections with empty tables so a re-run on the same corpus produces identical state (deterministic merge).
 	-- This is safe because M.run is the only writer to these tables within a single orchestrator invocation.
@@ -3255,6 +3405,8 @@ local function merge_corpus_registries(corpus)
 		"reg_use_schemas",
 		"reg_use_errors",
 		"tape_chains",
+		"atom_bundles",
+		"tape_emits",
 	}) do
 		corpus[key] = {}
 	end
@@ -3362,6 +3514,55 @@ local function merge_corpus_registries(corpus)
 			for _, chain in ipairs(scan.tape_chains or {}) do ---@type integer, TapeChain
 				corpus.tape_chains[#corpus.tape_chains + 1] = chain
 			end
+			for _, emit in ipairs(scan.tape_emits or {}) do ---@type integer, TapeEmit
+				corpus.tape_emits[#corpus.tape_emits + 1] = emit
+			end
+
+			-- atom_bundles: keyed by typedef name. First-wins per name (like components).
+			for name, bundle in pairs(scan.atom_bundles or {}) do ---@type string, AtomBundle
+				if corpus.atom_bundles[name] == nil then
+					corpus.atom_bundles[name] = bundle
+				end
+			end
+		end
+	end
+
+	-- Join slot roles to AtomBundleEntry_ identities after catalogs and atoms exist.
+	-- Do not invent a catalog from an entry that has no typedef.
+	for _, bundle in pairs(corpus.atom_bundles) do ---@type string, AtomBundle
+		local entries = {} ---@type table<string, string>
+		for _, slot in ipairs(bundle.slots or {}) do ---@type integer, string
+			local atom_name = bundle.name .. "_" .. slot ---@type string
+			if corpus.atoms_by_name[atom_name] then
+				entries[slot] = atom_name
+			end
+		end
+		if next(entries) then
+			bundle.entries = entries
+		end
+	end
+
+	-- Resolve tb_emit names: atom first, else unique catalog slot → entries[slot].
+	-- Ambiguous slot or missing entry: leave the raw ident (no A/B later).
+	for _, emit in ipairs(corpus.tape_emits) do ---@type integer, TapeEmit
+		if corpus.atoms_by_name[emit.name] == nil then
+			local slot = emit.slot or emit.name ---@type string
+			local hit  = nil                    ---@type AtomBundle|nil
+			local n    = 0                      ---@type integer
+			for _, bundle in pairs(corpus.atom_bundles) do ---@type string, AtomBundle
+				for _, s in ipairs(bundle.slots or {}) do ---@type integer, string
+					if s == slot then
+						n = n + 1
+						hit = bundle
+						break
+					end
+				end
+			end
+			local ident = hit and hit.entries and hit.entries[slot] ---@type string|nil
+			if n == 1 and ident then
+				emit.name = ident
+				emit.slot = slot
+			end
 		end
 	end
 end

scripts/passes/static_analysis.lua

@@ -4016,6 +4016,66 @@ end
 -- Each check is one table row and one `check_*` function.
 -- This is the plex pattern: the iteration is in ONE place (validate), the variation is in DATA (this table).
 
+--- @param src      SourceFile
+--- @param pipe_ctx PassScratch
+--- @param findings Finding[]
+--- @return nil
+local function check_tb_bind_type_match(src, pipe_ctx, findings)
+	local atoms_by_name = pipe_ctx.atoms_by_name or {} ---@type table<string, AtomEntry>
+	local info_by       = {}                           ---@type table<string, AtomInfoEntry>
+	for _, info in ipairs(pipe_ctx.atom_infos_all or {}) do ---@type integer, AtomInfoEntry
+		info_by[info.atom_name] = info
+	end
+	for name, info in pairs(pipe_ctx.info_by_atom or {}) do ---@type string, AtomInfoEntry
+		info_by[name] = info
+	end
+	for _, emit in ipairs((src.scan and src.scan.tape_emits) or {}) do ---@type integer, TapeEmit
+		if atoms_by_name[emit.name] then
+			local atom_binds = info_by[emit.name] and info_by[emit.name].binds ---@type string|nil
+			if emit.binds and atom_binds and emit.binds ~= atom_binds then
+				findings[#findings + 1] = {
+					atom  = emit.name,
+					line  = emit.line or 0,
+					check = "tb_bind_type_match",
+					kind  = "error",
+					msg   = string.format("tb_emit '%s' binds %s but atom_bind is %s"
+						, emit.name, emit.binds, atom_binds),
+				}
+			end
+		end
+	end
+end
+
+--- @param src      SourceFile
+--- @param pipe_ctx PassScratch
+--- @param findings Finding[]
+--- @return nil
+local function check_tb_bind_required(src, pipe_ctx, findings)
+	local atoms_by_name = pipe_ctx.atoms_by_name or {} ---@type table<string, AtomEntry>
+	local info_by       = {}                           ---@type table<string, AtomInfoEntry>
+	for _, info in ipairs(pipe_ctx.atom_infos_all or {}) do ---@type integer, AtomInfoEntry
+		info_by[info.atom_name] = info
+	end
+	for name, info in pairs(pipe_ctx.info_by_atom or {}) do ---@type string, AtomInfoEntry
+		info_by[name] = info
+	end
+	for _, emit in ipairs((src.scan and src.scan.tape_emits) or {}) do ---@type integer, TapeEmit
+		if atoms_by_name[emit.name] then
+			local atom_binds = info_by[emit.name] and info_by[emit.name].binds ---@type string|nil
+			if atom_binds and emit.binds == nil and (emit.data_words or 0) == 0 then
+				findings[#findings + 1] = {
+					atom  = emit.name,
+					line  = emit.line or 0,
+					check = "tb_bind_required",
+					kind  = "error",
+					msg   = string.format("tb_emit '%s' has atom_bind(%s) but no tb_bind_ or tb_data"
+						, emit.name, atom_binds),
+				}
+			end
+		end
+	end
+end
+
 local CHECK_RULES = { ---@type CheckRule[]
 	{ name = "transfer_hazards",                per_atom   = check_transfer_hazards                },
 	{ name = "gte_input_latch",                 per_atom   = check_gte_input_latch                 },
@@ -4040,6 +4100,8 @@ local CHECK_RULES = { ---@type CheckRule[]
 	{ name = "binds_no_substruct_deref",        per_source = check_binds_no_substruct_deref        },
 	{ name = "component_self_consistency",      per_source = check_component_self_consistency      },
 	{ name = "atom_calls_inferred_traffic",     per_atom   = check_atom_calls_inferred_traffic     },
+	{ name = "tb_bind_type_match",              per_source = check_tb_bind_type_match              },
+	{ name = "tb_bind_required",                per_source = check_tb_bind_required                },
 }
 
 -- ════════════════════════════════════════════════════════════════════════════

scripts/ps1_meta.lua

@@ -77,6 +77,8 @@ local PASS_FLAG_DISPATCH_KEY   = "__pass__" ---@type string
 --- @field atom_phases             table<string, AtomPhaseGroup>
 --- @field word_counts             WordCounts
 --- @field components              table<string, Component>
+--- @field atom_bundles            table<string, AtomBundle>|nil
+--- @field tape_emits              TapeEmit[]|nil
 --- @field collisions              CorpusCollision[]
 --- @field resolver                SourceResolver
 --- @field component_atom_infos    AtomInfoEntry[]|nil
@@ -373,8 +375,7 @@ COMMON_FLAGS:
   --help                Show this help and exit
 
 EXIT CODES:
-  0  All requested passes succeeded
-  1  Validation errors found
+  0  Ran. Findings print on stderr and in the report; they do not fail the process.
   2  Metaprogram internal error
 
 EXAMPLES:
@@ -665,6 +666,8 @@ local function build_ctx(args)
 		atom_phases             = {},
 		word_counts             = {},
 		components              = {},
+		atom_bundles            = {},
+		tape_emits              = {},
 		collisions              = {},
 		resolver                = resolution.resolver,
 	}
@@ -816,8 +819,7 @@ local function main(argv)
 		local requested = args.requested_set           ---@type string[]
 		local closed    = topo_sort(PASSES, requested) ---@type string[]
 
-		local had_errors = dispatch_passes(ctx, closed) ---@type boolean
-		if had_errors then os.exit(EXIT_VALIDATION_ERRORS) end
+		dispatch_passes(ctx, closed)
 	end)
 
 	if not ok then