From: Chris Duncan <chris@zoso.dev>
Date: Sun, 5 Jan 2025 22:40:10 +0000 (-0800)
Subject: Inline 64-bit addition to eliminate a function call.
X-Git-Url: https://zoso.dev/?a=commitdiff_plain;h=c2bf49af1a8bc8a5472b395efed25a78e2cd44cc;p=libnemo.git

Inline 64-bit addition to eliminate a function call.
---

diff --git a/src/lib/workers/powgpu.ts b/src/lib/workers/powgpu.ts
index e3a8518..67b5105 100644
--- a/src/lib/workers/powgpu.ts
+++ b/src/lib/workers/powgpu.ts
@@ -92,23 +92,44 @@ export class PowGpu extends WorkerInterface {
 		* LSb is the Least-Significant (32) Bits of b
 		* MSb is the Most-Significant (32) Bits of b
 		* If LSb overflows, increment MSb operand
+		*
+		*	fn add_uint64 (v: ptr<function, array<u32, 32>>, i: u32, LSb: u32, MSb: u32) {
+		*		var o0: u32 = (*v)[i] + LSb;
+		*		var o1: u32 = (*v)[i+1u] + MSb;
+		*		if ((*v)[i] > 0xFFFFFFFFu - LSb) {
+		*			o1 = o1 + 1u;
+		*		}
+		*		(*v)[i] = o0;
+		*		(*v)[i+1u] = o1;
+		*	}
 		*/
-		fn add_uint64 (v: ptr<function, array<u32, 32>>, i: u32, LSb: u32, MSb: u32) {
-			var o0: u32 = (*v)[i] + LSb;
-			var o1: u32 = (*v)[i+1u] + MSb;
-			if ((*v)[i] > 0xFFFFFFFFu - LSb) {
-				o1 = o1 + 1u;
-			}
-			(*v)[i] = o0;
-			(*v)[i+1u] = o1;
-		}
 
 		/**
 		* G Mixing function
 		*/
 		fn G (v: ptr<function, array<u32, 32>>, m: ptr<function, array<u32, 16>>, a: u32, b: u32, c: u32, d: u32, ix: u32, iy: u32) {
-			add_uint64(v, a, (*v)[b], (*v)[b+1u]);
-			add_uint64(v, a, (*m)[ix], (*m)[ix+1u]);
+			var o0: u32;
+			var o1: u32;
+
+			// add_uint64(v, a, (*v)[b], (*v)[b+1u]);
+			o0 = (*v)[a] + (*v)[b];
+			o1 = (*v)[a+1u] + (*v)[b+1u];
+			if ((*v)[a] > 0xFFFFFFFFu - (*v)[b]) {
+				o1 = o1 + 1u;
+			}
+			(*v)[a] = o0;
+			(*v)[a+1u] = o1;
+
+			// add_uint64(v, a, (*m)[ix], (*m)[ix+1u]);
+			o0 = (*v)[a] + (*m)[ix];
+			o1 = (*v)[a+1u] + (*m)[ix+1u];
+			if ((*v)[a] > 0xFFFFFFFFu - (*m)[ix]) {
+				o1 = o1 + 1u;
+			}
+			(*v)[a] = o0;
+			(*v)[a+1u] = o1;
+
+
 
 			// v[d,d+1] = (v[d,d+1] xor v[a,a+1]) rotated to the right by 32 bits
 			var xor0: u32 = (*v)[d] ^ (*v)[a];
@@ -116,7 +137,18 @@ export class PowGpu extends WorkerInterface {
 			(*v)[d] = xor1;
 			(*v)[d+1u] = xor0;
 
-			add_uint64(v, c, (*v)[d], (*v)[d+1u]);
+
+
+			// add_uint64(v, c, (*v)[d], (*v)[d+1u]);
+			o0 = (*v)[c] + (*v)[d];
+			o1 = (*v)[c+1u] + (*v)[d+1u];
+			if ((*v)[c] > 0xFFFFFFFFu - (*v)[d]) {
+				o1 = o1 + 1u;
+			}
+			(*v)[c] = o0;
+			(*v)[c+1u] = o1;
+
+
 
 			// v[b,b+1] = (v[b,b+1] xor v[c,c+1]) rotated right by 24 bits
 			xor0 = (*v)[b] ^ (*v)[c];
@@ -124,8 +156,27 @@ export class PowGpu extends WorkerInterface {
 			(*v)[b] = (xor0 >> 24u) ^ (xor1 << 8u);
 			(*v)[b+1u] = (xor1 >> 24u) ^ (xor0 << 8u);
 
-			add_uint64(v, a, (*v)[b], (*v)[b+1u]);
-			add_uint64(v, a, (*m)[iy], (*m)[iy+1u]);
+
+
+			// add_uint64(v, a, (*v)[b], (*v)[b+1u]);
+			o0 = (*v)[a] + (*v)[b];
+			o1 = (*v)[a+1u] + (*v)[b+1u];
+			if ((*v)[a] > 0xFFFFFFFFu - (*v)[b]) {
+				o1 = o1 + 1u;
+			}
+			(*v)[a] = o0;
+			(*v)[a+1u] = o1;
+
+			// add_uint64(v, a, (*m)[iy], (*m)[iy+1u]);
+			o0 = (*v)[a] + (*m)[iy];
+			o1 = (*v)[a+1u] + (*m)[iy+1u];
+			if ((*v)[a] > 0xFFFFFFFFu - (*m)[iy]) {
+				o1 = o1 + 1u;
+			}
+			(*v)[a] = o0;
+			(*v)[a+1u] = o1;
+
+
 
 			// v[d,d+1] = (v[d,d+1] xor v[a,a+1]) rotated right by 16 bits
 			xor0 = (*v)[d] ^ (*v)[a];
@@ -133,7 +184,18 @@ export class PowGpu extends WorkerInterface {
 			(*v)[d] = (xor0 >> 16u) ^ (xor1 << 16u);
 			(*v)[d+1u] = (xor1 >> 16u) ^ (xor0 << 16u);
 
-			add_uint64(v, c, (*v)[d], (*v)[d+1u]);
+
+
+			// add_uint64(v, c, (*v)[d], (*v)[d+1u]);
+			o0 = (*v)[c] + (*v)[d];
+			o1 = (*v)[c+1u] + (*v)[d+1u];
+			if ((*v)[c] > 0xFFFFFFFFu - (*v)[d]) {
+				o1 = o1 + 1u;
+			}
+			(*v)[c] = o0;
+			(*v)[c+1u] = o1;
+
+
 
 			// v[b,b+1] = (v[b,b+1] xor v[c,c+1]) rotated right by 63 bits
 			xor0 = (*v)[b] ^ (*v)[c];